Internet Engineering Task Force 102, 17 Jul 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF102-MBONED-20180717-1330

Description

MBONED meeting session at IETF102
2018/07/17 1330

https://datatracker.ietf.org/meeting/102/proceedings/

A

B

We can't fly this plane until we know this plane will not take off until we have a minute.

C

I go I, have audio processing delays and I can't fast enough. Mike Mike I.

B

D

That's easily Oh.

B

Are we expecting any more? Should we.

A

B

Wait another minute it's so.

D

You can talk about.

B

Not hey guys, we've we've got plenty of time. This time, yeah, it's a it's a small room. We walked by it a few times. We'll give him time.

B

We have to remote presenters. We've got about 90 minutes of agenda in a two two hour time frame, so we can wait a moment. You guys want to make some recommendations as to restaurants. You've been to that are excellent or bars.

A

You know I.

E

Mapped that- and it showed it was like four times- standard love, there's like way busy, so I want.

A

Well, if we do this post session here, but I like that idea, is.

B

That the one it's the bar, it's it's, the brewery, that's kind of a block away. Cuz I went to one of those okay.

E

B

I just went for lunch to some place and it was a brewery, but it was not a very wide selection, lowing.

E

De bruit, oh yeah yeah, it's.

A

Okay, okay, you know these typical bar very audible crowd, which I was it.

A

B

Was three brass presses.

B

A

B

Some interesting.

A

B

You're a lager drinker, it was not not a great yep.

E

Thank you thanks.

B

She wasn't here to tell us that we're supposed to get her started for real.

F

B

Thank you. We appreciate you being a note-taker as always you're the best note-taker we've ever had.

E

We got one miking why this volunteer, okay and I want to make that publicly know that we, we really appreciate you in every way.

B

All right, I think we should start now start Jeffries here, so we can start all right. Welcome to PIM so welcome to M Bundy. This is M Bundy. uh Here is the note. Well, please note it well folks, on the Medeco we see Dale Jake, Kerry, Lorenzo, Lukas and Natalie. Can you guys hear us.

B

Yeah, this is the widest audience. We've ever had remotely, at least for a while.

B

That is a yes okay from any of from those people. Okay, great cool, all right. Here's our agenda, we're going to switch things around slightly!

B

You want to fullscreen that.

G

B

We'll get there so this is the agenda a fairly full agenda. It's a good thing. We got our own room now because we're gonna we're gonna use it our own session and oh yeah, blue sheets. That's a good point.

B

Yeah! These are the blue sheets, be sure to sign them.

D

B

Right, please sign the blue sheets Toshi's here to talk about em trees and he has some late-breaking news that we're excited to hear about. I'm sure mike has a couple presentations on update and new beer. Entropy Jake's gonna be presenting remotely we're probably going to move Natalie up right before Jake to promote true Pro to present remotely and then toyless has a couple presentations.

B

So, let's see next slide, alright. So of our active working group documents, we have em trace, which has been stuck for a while with security issues at is G and huto she's going to provide an update. Hopefully he has figured out a way to unstick those.

B

We have the deploy draft that they're the data center deploy draft Mike's going to provide an update on that Mike. You want to tell us: what's going on on the EM cast problems, problem statement, draft.

H

Like my bride, we didn't do any update since the last ITF on that a I Triple E 802 multicast problem statement. So we didn't ask for a slot this time, so I'm not exactly sure. Probably what we'll do is we'll probably do another update before or Bangkok represent it. There I think will probably be pretty close at that time. To ask for a working group last call cool.

B

And then the yang models draft was adopted successfully since London, so sandy is sandy here. No, no sandy, that's ready to be. You know, reissued as a working group document. So after a while you know took a while, but now we're a working group document, so we're ready for that other drafts there's the deprecated draft, which is actually in a current ongoing call for adoption. So we will. We would really appreciate folks replying to that adoption call either yea or nay, if you think it should or should not be adopted.

B

Toilets is going to provide an update on that one. There is the am bi draft that Jake is going to provide an update that you're just going to present, tell us all about and then there's the beer entropy data center draft that Mike wrongs are going to present.

B

Okay, so first up a Toshi.

B

Now I did post your latest slides, but I also posted the latest slides in the agenda slide, and it wasn't the latest so it's possible that it didn't get updated so be prepared for it not being their latest slides, I.

E

B

Do the M trace one yeah, it's missing stuff: let's go to the other web page where's, the data track; click on the m bundy agenda, one; the air attack yeah.

D

B

Let's click on that about not do anything.

B

No, that's not it. Can you go back, go click back on the agenda.

B

Yeah but just try it, it.

E

B

Right, how do you get this page.

B

Yeah but it might.

E

The working group, or they I,.

I

Stupid working group chair question what M bone D? Actually, you know, do the work for any operational document related to beer.

J

I believe it would, it would apply I, don't think, there's an inch anything or charter that proposed.

I

We haven't seen anything yet right so operation.

J

You haven't had in the ops requirements, I mean if vision will happen. It's part of the discussion came up in like what does beer do now that it's not just trying to stall here's. It's fine, no.

B

Need the stall no.

J

But that's interesting.

B

Conversation, that's actually a good question about.

B

Mike there is a good question that maybe you can give us a at the end. What is the state of beer and are there operational considerations that we should be providing here.

A

I'll shoot I.

B

Know she I think this. Is you.

A

F

K

No okay, so this is a history of status, so we studied the ISD lay beyond generally 20 so this year and the budget was 22 and we got several comments from is G and we revised the based on comments and submitted 20 version 23 on appeal to others old comments, but mainly ma normative wording or irony related things and so on.

K

So except media and Alec all approved I believe, but we had a still comments from media and alec and we leave ice based on their comments and submitted a version 24 on June to August their comments and media upload about the eloquent. So we have been working to fulfill the requirements. Eric showed and so today, I just briefly summarized status and the current situation and they go to the next. But beef maybe just go to the last Friday's weather. The last strike last try first.

K

So actually this is update on this at this morning. So we thought after Eric's comment confirmation that blue ball. We will submit new new vision, 25 and ask final procedure, but I met him this morning and finally, I got his confirmation of a blue ball, so now I'm think I believe we can submit to the worst version and the desk device body will be approved by all ISD.

K

So this is a current news, but so if there is no consensus- and there is no appealable I may want to ask you how we can address this comment and how we can stack the current situation, but luckily I got a consensus and approval. So actually, as a result, maybe I don't need to ask what should I do now, but anyway, I can summarize what we have been working. So in a previous ride, you have some question or comment. Well, no.

L

I was just gonna: ask you if you would say something about what you had to change, what was needed sure.

K

Sure, of course yeah. So this is just a result. Iii think it's better to show the final result, but now I think it's yeah yeah, it's great. So, but let's go back to the page II. Maybe you removed all the suspense.

J

K

3, yes, this one, so the his comment, mainly his comment has he has two comments. Why is forgery of responses? And so it's mainly security related stuff forgery of responses, amplification attacks- and he said this product- does not appear to verify that the sender of the request actually owns iCal claims.

D

K

Anyway, so no keep cubed yeah this one and because the responses are much larger than the queries. This allows for application attack, especially India. If the client is able to send a query request study it's much reprise and there because a query IDs so short an attack and can generate first.

K

Okay, well, so he required something like a longer qle ID, but we already replied that the clearly ID is not intended as a security protection mechanism. So the length of the query ID is meaningless to protect something from the security slits. So the first part is actually the main point, so we updated based on his comment. We updated the forgery of less horses so well. He may acquire something like encryption mechanism for the messages, but we don't want to adapt to some special encryption mechanism.

K

Maybe it can be, but that kind of specification could be specified in a different documentation. So we don't want to say the encryption making me something like a master assured. So we explained that anchor fix make increment. It could be fine, but that kind of party is out of scope of this document and he understand this part. The second part, this is more sensitive, so filtering of clients so see this one is a version 24. This includes that bird. This sections have a section is included in version 24 and also in 25.

K

We try to modify the wording from 24 the letter. Let fonts explain the differences from the 24 24 and the answer that apply from us is that we should specify some configurable packet. Filtering mechanism is master so because maybe it's better to explain with this example. So let's say we have one client and this client is attached to allowed a and Lauda is upstream with Delta, B and larger C and so on, and at the letter C attached to the sender. In that case, allow client sent actually messages legally.

K

So he is a very good member. He sent a clear message and Lala I accept his query and translate this query as a new request message. So the quest message is transferred from A to B and then B to C and then C, Li change, the header part or the message type from the request tool apply and send back to the client. This is ordinary case, but his concerns.

K

If the body receiver is attached to, let's say, luta, flautas B and the Delta, B, intentionally or unintentionally sends a request packet, not actually packet, request packet, but louder B only uploaded his IP address. So he sent that this request message to assume louder's, but in that case request message may have some problem, so example over the empty you or something, then the LMS egde is transmitted to the original client, which is attached to Alta a so. This is something like a amplification attack. He said so this this is amplification attack. He says so.

K

This kind of attack or intentional or unintentional attack cannot be, cannot be protected by any mechanism which is because by any previous mechanism, but to protect this kind of attack we specify configure packet filtering mechanism is: must this means that the router should also recognize a type value as well as IP, address and port number, then the loudest thing that this request message must not be given by his downstream nose, even though he is a valid member, this kind of intentional or unintentional message must be dropped at a lotta, B, otherwise, louder.

K

B just believes his request message, which must be different and it must be strange and the transferred transferred to the upstream router. So, to protect this kind of attacker, we should provide some configure packet, a filtering mechanism that supposed to verified IP, address port number and also the type value of the messages.

L

So does the filtering need to happen only in the actual entries, client or locking the entries code like the processing code? Are you saying that the router this condo just on the path to the IP destination, they'll need to do some filtering.

K

Second, yeah.

L

Basically, I'm trying to understand: okay, it's it just like when they're out the processes and em trace packets, like as part of being a mem, trace, client or server whatever that you can do this filth Rima. Do you expect like a router that forwards IP packets to see? Oh, this is an MJ's packet. I need to maybe drop it well.

K

The client, which means and the hosts only can initiate it clearly not to request or.

M

Reply: okay, but.

K

Let's say: there's some body door embodied the client, which is just a house not a louder, may send a reply or request message. This is something a bogus.

K

This is something like attack, and in that case the problem is let's.

K

So this is a little bit complex, so the if the clearly the leak quest- let's say, okay, so the first does send the clients and actually and clearly sent curity packet is now translated to this request by louder, and this request is transmitted to other sender, okay, but in that case the requests include, add the appended append response block, so the packet size will be increased by hapa and in that case, in the middle of the pass, the packet may encounter the MTU limitation.

K

In that case, the current packet request packet is transmitted to the original clients and continuously eliminate by eliminating the previous standard response blocks. Only keeping the message header, which, whose type is a request, continuously transmitted toward the center, but his concerns. If the message itself is really larger or the empty which service is really small along the past, then the an incomplete Leafly message may be transmitted by each louder several times.

K

So, even though the the client send a one single query message, he may receive the multiple replies which is initiating it, which it is created by intermediate routers. This may so. This is an amplification attack. He there. So to avoid this kind of attack, intentionally or unintentionally, attacker request message must be.

K

Originated by intermediate louder, not any host. Okay,.

L

So I think them basically only implement and traces part your entries of processing code. You have to implement some kind of check to see. Yes,.

G

Yes, yes, exactly.

K

L

If it's just some random rafter that just see this.

L

Thank you, my my concern is you know when you say like packet, filtering and stuff, it kind of sounds like it's like an HDL that drops the packet like you know, and it's not on the wire, but this is actually in the embrace. Call that well.

K

But this is actually really similar to the ordinary, regular access control, ACL mekinese. The difference is that ACL only support to filter the IP address and port number, but as well. The message type must be also checked.

L

My concern is, you know, don't think you would get all all the offers on internet to have the capability of filtering packets based on this is.

K

His requirement, but.

L

If someone implements and tres versus two, then they can certainly implement code that can check. Okay, it's the 500 small K address right. So is it sufficient to do the letter thing that the only the routers have implement MJ's person to need to, but.

K

I heard it from Kelly's mountain no I said.

B

K

Okay, so he said there is some Ben. Does implementation on corresponding to this kind of a filtering stuff? She score juniper.

B

K

He may say something: Carrie.

B

If you want to speak.

N

There something we need to do grant.

L

Is like hey I can.

J

Just click anything just keep.

A

L

K

Anyway, he said there is such kind of an implementational, the something like a common implementations to filter out the based on the message type or something so that cam. That kind of filtering mechanism can be at a potato for this kind of a schemer.

L

So it might mean that you need to upgrade all the routers with new code. Well,.

M

L

M

Possibly exhaustively.

L

You know possibly.

K

But anyway, so this is and also.

K

Okay, so okay! So anyway, if you don't ugly this kind of sentence, please say the mailing list or.

B

K

B

Eric required in order to get okay, yeah and stig you're you're, not agreeing you're. Can you have concerns that it requires all the routers to upgrade, but really that the routers have em traced be to working now anyway? Yes,.

L

It's in in practice, I mean you could say: people must implement some filtering to support. That is that right to be compliant with whatever they are disability, and you can say: okay, if you support your people, implements em chase, you need to be compliant, of course, but it's you know it's hard to enforce and say that wall routers that are deployed need to support and trace filtering, and if that is needed, to make it secure and in practice it. You know, you can't really expect that to happen.

B

Okay, but this is em, trace v2 and do does em trace v2, don't they need to upgrade in order to support this I.

L

Mean you need to upgrade when you are doing the trace each router on the path with thus beginning of the trace towards a destination need to support and trace, but.

M

L

Routers than between the clients and those brother is still so, you know they will follow the packet as well. Here like like the curry or the respond house normally do they need to support children, but.

B

It's for taking that to the list just so we can continue on.

B

How did they say how to grant him the floor.

B

It's a great big red button, but it's not labeled.

C

B

K

You shouldn't use the.

B

K

N

B

Carrie, you can speak hi yep.

N

So there's there's standard.

M

N

That's I'm, sorry with someone just talking: okay, there's standard functionality that is supplementary to ACLs, that's provided by juniper and Cisco and I assume other vendors that provides generic packet filtering and it allows specification of an arbitrary field within the payload which in this case, could be the M trace message type. So this isn't a special thing required as part of the M trace implementation. It's just there for current releases of a number of vendors routing and switching products.

I

Taught us Eckert, I, guess arbitrary within the data wouldn't even be necessary right. I hope that the message type is some: when the first, what is it 32, 64 bits or so of the message? That's to be you what I remember everybody supports yeah.

N

That's right, Troilus, so it's within that arbitrary, but some number of bytes of the header, which in this case will work fine. We can differentiate, request and and reply from other message types. So we can say only these ranges of IP addresses are allowed to send a request. And/Or a reply message so.

L

Carry just run this, and this is speak again, so so it's not sufficient to filter this. When the process I carry our response, you actually need to forward. Okay, yes,.

N

So it's a a per interface required configuration on routers that are being used for M trace on.

L

The roughest race, correct.

N

So the configuration, although the you know of functionality, would I assume it's there on all the routers in the network. um This filtering only needs to be configured on routers that will be participating in M, trace. Okay,.

L

That's great, but doesn't that mean, though, that just as part of normal and risk processing, you can say all if this message come from an acceptable IP address.

N

So at first that's what we were thinking, so we were just gonna, provide a CL filtering, but there was a amplification attack scenario.

N

The Eric suggested that really requires us to additionally filter it by message type, which is that we have a allowed routing peer that has been explicitly enabled by IP address, but instead of sending a request I instead of I'm sorry, we have a client that is allowed to send request and the client.

N

Sends I'm sorry, it's all absent queries and the client sends a request. Instead, that.

L

That mean, though, that all you do is you know in there I'm just implementation, you just checked okay, this is like Carrie doesn't come from all of the yellow carriers or if it's a response or the complement, acceptable.

L

Yeah, whatever the Lords see what I mean when you process the empties message, of course you know what typed messages it is, so you can make that decision on.

I

Your main clarification, and maybe that helps with mistakes questions so try to remind me of these packets that we need to filter our guaranteed to be at the process level, because they've been hunted by router alert or something else or it can they be in the aqua forwarding thing and need to be filtered there. So.

N

We want to prevent them from even getting punted. The the reason for using the filtering mechanism is we'd prefer not to even interrupt the CPU with with disallowed packets, like this.

B

What we're quickly running out of time, we've eaten all the extra time that we had.

D

B

Yeah, so can we can we take some of these things to listen? Can you quickly go through the other two yeah.

K

B

Maybe summarize them because actually.

K

This is a most, so next one is just explanation about what is application attack, and this is also related to the such kind of filtering mechanism, so più itself, for example, TSM or pin D, and so on- has mechanisms to authenticate the neighbor louder's. So, in addition to the such kind of filtering mechanism, router this one says should top any empty edge to request all the premises that is received from a known authorized, be neighbors, and additionally, this is a today's agreement.

K

So this initially we didn't mention this kind of a scenario, because MTS doesn't rely on any kind of multicast routing protocol, but he requires such kind of authentication mechanism. As an example, we explained about a FEMA authentication mechanism, but not only p.m. we also need to explain the intent over this text is to prevent non louder in point from everywhere in village of non pimp. Loadable should employ some other we're going to prevent this kind of attack. So if you use gnome PM louder than something like a similar mechanism must be are considered. Okay,.

L

Stick again, I just want to say quickly, so you know it's great that he accepted this, and maybe we can, you know, be fine with it. I think we need to spend a little bit time off in the working with just to make sure that what actually was agreed on this practical implement so.

B

That the hostages have been agreed to be released, but we're not quite sure we agree with the terms of the release. Okay, yeah I think I think it goes without saying. We should definitely send things on list when people have a chance to really review and and and digest what was agreed to make sure that it didn't change anything okay,.

K

So we will submit anyway 25:25 and then after submission, then please leave you the us to version and.

I

K

I

Think at this stage in the process, right except I'll, propose something better, okay right. So yes,.

O

I

Concerns I think we're past that okay.

B

So hopefully I mean you know it's like they say. The 25th draft is the charm, so I.

J

B

J

B

And thank you Higashi for your grit and determination and getting through this process of 25 different revisions in a long journey.

H

Very good all right, so some of you may remember this draft from thank you from years ago. Last time, last ITF I mentioned that it was resurrected. It was already working group draft. We thought it would be good idea to have a working group draft here in mbone that discusses multicast as use in the data center last ITF I.

H

It was just me on the draft I requested someone to join me and that request was granted by olafemi from Arista, so he did a great job of fairly significantly updating this draft, with some new information and.

H

It's just this movie looks like that's all. There is anyway yeah, alright, so basically, what we added was things such as things that may not even be necessarily multicast related, but more one-to-many related whatever that may be. So we address things like hidden replication.

H

We introduced just a few things that in the future, could be used in the data center, and this is where we would need to get your feedback here, and this is something that was mentioned at the beginning in this meeting, and that is that we included just a brief comment about multicast and segment routing.

H

We just mentioned that in the spring working group, for instance, they are reach, are during, but multicast continues to not be included in that Charter, and so we need to have a place to address multicast in spring. We're addressing some of those things in PEM within embo Andy it'd, probably be a good idea to start at least thinking about some ramifications and in beer right, okay, very good, and so we were dressing that where we meant we mentioned beer in this draft as well.

H

Just you know what's what's being worked on, even though, as far as I know, there's no beer deployments than a data center is that a correct statement today would that be accurate? That.

H

You can talk about so.

B

That's not sure, so there are some clear deployments outside.

C

Of data centers.

B

So should we be doing things like a problem statement or a requirements Draft for spear and spring I mean for multicast and spring. Maybe.

H

So maybe so cuz we have. We have, for instance, NPM a giraffe, that's going to be mentioned today. We've had some that were discussed previously about different solutions for multicast in and segment routing, whether that's beer or some new type of solution. So maybe the some of those things may need to be discussed from a deployment perspective. Here.

H

L

A stick- yes, I was wondering a bit what you want to do with this crap to talk about. You know what people have deployed like this practices actually talk about. Also like issues people see or like future technology of phone installations.

L

H

Prior to this latest update from Femi, it was just what's been used today up to this point where March kids used, but we added a few just little tidbits that maybe we shouldn't of well just just an FYI. There are some new developments that are being used, they're coming into deployments, probably soon just be aware of them, and if, when that does happen, here's some of the basic things to be aware of.

H

If that's something that that and that's what we've added. So that's kind of the question: if we think that that kind of information could be briefly mentioned in this draft, as we've done or left out entirely.

L

H

Okay, I think we will focus on keeping it to what I just mentioned. That is current deployment of multicast and data centers. But again we can't help ourselves by at least mentioning some of these new technologies coming along, so yeah, so any way that we encourage people to look at it further.

H

We're not ready to ask for a working group last call we're going to continue to develop this we're in no hurry, but I think it could be useful, especially with this latest update and any questions about that. Further.

H

Anybody wants to join in on the fun feel free to join us.

H

This would be me yeah, maybe both.

H

H

Much for that crumble.

H

Just eat what your puppies, actually, one of them.

E

H

B

Charles, let's go back to this: let's try the hat kind.

H

Of the reviews that don't look for our meeting yeah, it's really helpful.

D

H

Just keep it on that first night. If you get yeah, not open.

H

Yeah all right: okay, yeah! If you actually go to the first slide, that would be helpful back there. Yeah.

O

H

Hopefully, you can see this so a few bets, including someone at Alibaba, thought it would be helpful per my previous comments to address beer considerations in the data center, as beer is getting more popular and as we discussed beer with customers in some cases, we need to convince them of the the benefits of beer and they understand you know the state, saving and all that, but we found that it would be helpful and I think that's part of the mbone charters to do these kind of deployment and consideration documents to just kind of help.

H

People have something to point to where this is what you'll get. If you do use beer- and in this case we we could either have a beer, general beer and a data center draft here in mbone d or beer, or we do like what we've done here and that is, we've just chosen a specific feature of beer and then, in this case it's including entropy and beer for load, balancing and just and create a document about it. Yeah.

J

Just just for technical clarity, what you did was you selected a use case in the data center and then looked at how to apply beer for that use case, and so and then just using that framework, that's kind of where mbone D belongs in a deployment mecca's for multi point services, but you're presenting this in beer as well tomorrow, correct just.

H

For kicks and giggles but yeah and.

J

You should I mean we normally do that across across working groups to make sure everyone's informed. Clearly, the entire beer group is not represented here. People might have issues about how the intro field is what I mean I I like where this is going. I, don't disagree with all the decisions, because I've got some some feedback into this as well. Take a stir: okay,.

H

That sounds good all right. So okay next slide the problem statements. So if we can get there, okay, very good, all right so I tried.

E

H

This draft described in Jing wrong, it's been leading the and these are his ideas and I've just been kind of helping them with some of the details of this, but we're describing how beer entropy can be useful in data center Clause networks, so beer, entropy, entropy in general, is being used with a variety of protocols, including segment routing. There's a draft on entropy I.

H

Don't know the RFC number yell it out if you know it, but that allows you to view load balancing and it can be done in a more deterministic way, rather than just more of a random hashing and ecmp kind of environments and beer includes the entropy as well, so you can actually have, as you look at a packet and look at the source, IP a destination address, and maybe the port numbers you can take that out and create an entropy label, an MPLS environment and hash. On that. Just tell the second sis question.

I

I'm trying to stay clear of data centers, there's so little oxygen, but any any example of kind of the worst case. Multicast in terms of you know, to think of an application where you know getting. This right is most important.

I

J

You have one no professional media, okay, we're like your description here, the Greg Sheppard Cisco, it's not chair, had you said long lived elephant flows in pmn, they're, all elephant flows, and so when you've got multi point going up, multi point come down. Ecmp could create collisions based on the addressing hash, whereas you want to make sure that this flow takes this link and the next big flow takes this link and there isn't collisions across that typically kind of streaming out of the data center or kind of the always.

I

J

All within the DC, so processing for exactly so you've got I mean so even well. What we call a DC and that could be in a truck, could be in a stadium. It could be in a massive data center facility, but the flows are the same they're all 4k, HD cameras and, and they stay that way. Uncompressed flows I think that the important.

I

Part that you mentioned is the data center doesn't have to be the Alibaba big thing right, but can can be arbitrary load, small and the small yeah you.

J

Basically, just you've got leaf in spine, so you know to hop you're trying to just do two hop deterministic ecmp paths, yeah.

H

That's exactly right exactly so that's and that's what a clause network is, and so without going into much of detail on that, and we mention it in this draft and I. Think I may mention it more in the DC deploy draft, and that is, you know how we used to, for those of us have been doing this for a long time. We used to have the hierarchical tree like structure, a tier 1 tier 2 to3, which was core distribution and access, and we've kind of moved away from that.

H

It's much more popular to do these cause this spine leaf scenarios where you've got to 1 to 2, to 3 the spine being tier 1, the leaf being tier 2 and the top of rack being tier 3, and it's more horizontally scalable. You can add new servers devices routers whatever to the spine as you need them without having to replace boxes. So it's just much more scalable and that's for port density. That's why we're doing it and we've! You know kind of mentioned the the long-lived elephant flows.

H

They may affect performance of the short-lived loads and reduce efficiency, and there are some hash function: inefficiencies in ecmp and you could have frequent flow collisions and more flows can get placed over one path over the other and isolating faults with multiple parallel paths in a data center.

H

Whether and there's a lot of them can be non-trivial due to lack of determinism and so we're trying to find we're trying to specify that beer with entropy can give us a level of determinism to help alleviate some of these problems and an operator may want to have a deterministic path, and this this is something that's being used elsewhere. There's a draft MPLS spring entropy label, which addresses very similar thing.

H

We've utilized some concepts from that draft, and so basically, a deterministic path can be found by using part of the 20 bit entropy build that 0 2 bit 2 friends, for instance, that the entropy level can be can represent the value of zero to seven. It can be used to select deterministic paths from a equal cost paths. So again it's something that's not new to beer, but thankfully beer does utilize it.

H

It's mentioned in the beer dress in not any particular great detail, and so in this draft we intend that if this is just a first stab at it, but we intend to provide some mode more that detail of how you could use this in a data center network. Any questions so far.

H

Okay, so I'm just gonna quickly go through the next two sides. It just kind of it just kind of shows what I just described in this, and these diagrams are used in the in the draft. So go back up one more great: it just went down yeah there we go stay there right there all right, so this kind of shows in clause networks, there's three stage: there's five stage architectures and you have northward stages and you've got southbound stages that have a rich amount of equal cost.

H

Multi pass I've already mentioned some of these problems for steering, steering elfin flows and path, divisions, two different sis, okay, next for yeah yeah, it's hard I, know I'm, not we're not going to use that.

H

And so again, by using the beer entropy, you can bill to provide more deterministic path, selection and you can use like I mentioned in stage. One, for instance, use bit zero of that twenty bit entropy label to represent red and blue plus, and you can on different stages. Let's this case stage to use different bits of that enter beef label to represent three paths to each of the different clusters: to be able to provide a deterministic, you could cost multi path along a data center, multi stage network next slide.

J

Yeah, cisco, so in this I your showing how how to potentially break up the interview field to set deterministic paths, because the question I have is around where the bier header gets written in a way to use these paths, so that would have to be I guess below stage one or it could happen either way. I mean she had IP multicast at the pod level and beer in the DC. It's the first in position. Point creates that field right, yeah.

H

That's that's correct, so in my view it would be in stage one it's a local decision, that's being made and so stage.

J

One being the pod so you're doing in position at the host level or you're talking about I guess.

H

It could be at that I wasn't thinking of host level. I was thinking in the pod, but so.

J

As I've been multicast from them from the host into the pods right now, the pod is aware of the topology decisions so I'm, just without taking away your Thunder here and going.

P

J

This could be a DC specific decision how the bits are used, where I could be just networks like beer domain decision not necessary. We have to have our architecture dock that specifies the bits to be used in this way, but here is a particular way to use those bits and how you cut them up. Based on how many CMP paths did a complexity for topology, you can set a beer wide specific configuration. You.

H

Should cut using controller or whatever tip that happened? That's right exactly right! Okay, so next slide yeah, okay! Well, we're just about done! We're done so go to slide six. It just kind of specifies some of the forwarding procedure using beer in tributed, just like the path over equal cause. Paths already explained at this draft defines a method to use part of that 20-bit entropy label, and it should be easier and should be more granular and more deterministic than using just the typical hashing ecmp function.

H

So with that I was just going to leave it at that and just kind of go on, but I think it would be good to have a discussion about if this is something that this type of weather it's just draft or not. This type of topic is that this is something that we think that we should address work on here in an Bodi question.

I

Yeah so I think you know the most simple, but you know problematic reference example to show you know they have people belief in the problem right so I think, starting with what you know, Greg was saying: maybe you know the smallest kind of data center right, which could be production truck with a bunch of you, know the small stuff and then video whatever. It is right so that dead set the expectation in terms of yeah. This is a problem and then the other things are fine. I mean I completely forgot the details.

I

So do we have polarization in the beer ecmp? That's one of the typical issues to deal with right, I, don't remember it in the algorithm. Obviously you know I also stand out on. You know behalf of this author of beer te and say that we haven't thought about really that much about the EC and P stuff. It was obviously mostly about non ecmp path, but we could also think about comparing this what we can achieve with the beer ecmp with the explicit hop-by-hop engineering of the path and the data center with beer te o your.

J

I

Okay, right so I mean I. Think there options I would start with the you know. Do you buy into the problem statement with you know the best possible reference example. We can come up with because you know if you buy into the problem statement what what else can we do then work on it right so yeah.

H

Okay, thank you any comments about. Well. Maybe this is a chair discussion, but this being worked on in beer or in FOD, so.

J

Gotcha, so the question is: is this what so who's read the draft.

J

Who feels this is material that should be adopted as we refine them in this work well or anywhere. Yeah. That's a good! That's a good question! Jeff reality, reference.

Q

Univer, if this involves defined methods of using the part of entropy layer failed in beer or the purely entropy label, were what the came first clarified. As you see on this slide, you mean you mentioned that gives part of the 20 bit entropy label. I talking about entropy label are talking about entropy filled in the beer header.

H

Do you enter that you run.

P

In top, you failed in the beer harder, of course, but generally speaking in entropy is followed, the with the label. It's 20 bit value. Just there 20 bit value.

Q

Okay, so if this involves pure entropy field and also involves some specific beer, falling behavior, then I think ish should be part of the beer working group in other comment is that if you want to control how the that of the past that the packet follows another way to do, this is use multi, topologies or or just a flexor I'll go so that you can decide that some Packers follow so one sub subt apology, some other Packers poles are differently sub topology. That's another way to do it right.

J

Can I'm gonna address that one in particular, because I think the use case that I'm thinking of that kind of came up from reading? The document is specific enough that the multi topology it would be very complicated. You would have to have several of them not just a couple and we're not really trying to select topology necessarily we're trying to ensure that each of the hops don't have collisions. So you could build complete matrix of topologies to ensure that's the case, but in a flow by flow basis.

J

You may want to make a different determination that this link was okay now, but now at the next hop down, we've got a congestion issue. I want this to keep on this hop, but take the next one over here to avoid that congestion. So it's that's something, but this matrix of topologies could get very complicated quickly, so I think to Tauruses point. Maybe a more specific problem statement would allow to burrow into a solution that is very specific to that problem.

L

That's dig I agree that this should be done in beer I, think because it does seem to alter the ear forwarding and some rate right.

B

So you're presenting it's tomorrow and beer yeah for more than just courtesy, no.

J

I

So TARDIS it wasn't kind of clear to me. You know if you're really changing something in beer or just basically, you know what a application using beer and setting the field would be doing so might be borderline, whether it should be beer or in Bondi, but overall, the problem statement. You know describing the example reference problems, so anything operational would be good here, so I mean I, have no strong opinions either way right.

A

L

Don't remember now, does the beer specifications say how to set the entropy exactly how it should be calculated or is it left more left, open, I.

H

Think it says it's not a scope in that in the document, but we should double-check that. But.

L

If it doesn't, you know required changes in beer forwarding it's more obvious, any of them yeah.

J

E

J

How the value is set versus specifying well, it's kind of both really, but instead of the entropy field, pointing to a nice EMP table, it points to some deterministic table. That's effectively ecmp, but it's not built on hash. It's built on some predefined topology yeah.

B

So yeah we're running out of time, but did we need so you're just gonna present? So why don't you a.

M

B

And then see the good news is we have a common co-chair.

E

B

Both so Greg, you know you can kind of lead the charge in determining.

H

Where this belongs, thank.

B

You all right next up, Jake I, hope you don't mind. We're gonna, we're gonna move Natalie up ahead of you and she's going to present on.

B

Her experience in AMT Gateway implementation building in VLC, so Natalie are you there? Yes,.

M

Can you hear me alright, yep.

B

So, just in in London we had William Jiang present. His experience in deploying an AMT relay one of his pieces of feedback was that okay, the relay is up and running. There is content out there. However, the state of Gateway implementations is not so awesome. You know the best most widely usable implementation was a VLC Windows port that was about 10 to 12 years old.

B

So natalie has been working on examining that the best strategy for deploying gateways and she's going to talk about her experience, what led her to VLC and what how things have been going.

M

Alright, excuse me um so yeah. Let me explain the background for that. A lot so um I am you can probably just go to the next slide then.

M

So the end goal of the project was to allow end users to access multicast content over the Internet, because it's out there, but there's not an easy way to get it. um You kind of have to do a lot of workarounds and so I made a list of various ways that we could do this so that in the end it would be easy for even like my mom to use whoo she kind of struggles with technology. um Cuz I don't want it to be.

M

She has to enter an IP address and like go through all the stuff, because she, you know something goes wrong with her email and she's, like I'm done, I quit so I've been doing that since the end of May, that's been my project at juniper and I had no idea what AMT was before this. So it's definitely been an experience.

M

You can go to the next slide. So again, I made a list of a bunch of possible platforms. Our original goal was to use a web browser implementation because that tends to be the easiest for people. I mean you just go to a website, you, like click and it's there. That's really easy for most people understand and use, but I did find some issues with that.

M

So um with web RTC, at least from what I found it had no multicast support and then webassembly had no UDP support, because amt works over UDP without that it just doesn't work. um So then I looked into making a plugin for a browser like Chrome Firefox, and then you know making it available all of them. But I found out that google apparently deprecated plugins, starting in like 2016.

M

They have that had this like movement and then it kind of like finalized in 2018, where plugins you can no longer create new ones, even though older ones can still be maintained, so I didn't want to start on making a plug-in just to have it be no completely out of date. Soon after and other issues arise with that, and then I tried an extension and the gist of it was that anything to do with the web browser seemed there's no raw UDP access. So um I'm sorry was there garbled speech, nope.

E

M

M

I'm sorry I, don't know why it's a curly.

M

B

Not hearing an echo if, if that helps.

M

Okay, um but with the browser implementation, there used to be an API for accessing UDP sockets, but it has been put out of date or like when I even just started on some of the tutorials. For, like writing a plug-in or something like that, I kept getting all these errors about like hey. This is deprecated you shouldn't use it like try moving, is something else, but it's newer. So um there's there's not a an API and it's understandable from a security standpoint just accessing raw UDP sockets, but they're.

M

Also Google is pushing progressive web apps, which then it seems like Mozilla was following suit from the stuff that I read and there's again still no UDP support, so I settled on VLC integration, because VLC II is essentially the most widely used media player that isn't the default, because if you there are a lot of people when they have issues with their media player, if they go to Google and search for something, VLC is like the next thing there, and it also it already has native multicast support and it's available only all platforms like they.

M

They have a port for basically everything and it's backwards, compatible for, like all their old versions and stuff, like that. So uh and theoretically, I actually found this out more recently. They also have a plugin like that has already been made. So theoretically, if the amt module got added in then you could make a web application that could use that plug-in that as long as everything gets flushed up straight and such so, you can go to the next slide.

B

Still there Natalie Oh.

M

Can you not hear me you've.

B

Kinda trailed off there, the last 10 seconds, oh I'm,.

M

Sorry, what was the last thing I said.

B

We didn't hear it. Oh sorry, we're you, probably you ready to go to the next slide.

M

Yeah, okay, hopefully this um works out better I I think it's even.

L

B

Bike but I'm not sure what.

M

Is it better? Yes,.

B

Yeah now it's good okay.

M

Cool so the other benefits of VLC it handles basically all major media types. It can play almost anything even like corrupt files, which is really good. It's also modular, so adding parts into it and like developing on its source car source code is pretty easy because you don't have to worry about breaking it. The other functionality too. Much and again, it runs on almost any platform. I found that it's pretty easy to use from my standpoint, but then again I'm also, you know computer science person.

M

It could possibly be easier to use for people who don't, but it still got a lot of instructions out there in terms of how to use it. So you can go onto the next slide.

M

um So what I did is I started on an access module. They VLC has all these kinds of modules that work for different things like output, access, D, muxing, so most of the D mixers are actually the what decodes the streams and stuff like that. um So my idea was that all create an access module that, essentially you know open supports, sends the correct messages, the discovery advertisement and such and then pass that data onto a d'emic, sir, because theoretically, VLC can already read whatever the data is because it should be a major media type.

M

It shouldn't be something ridiculously obscure. I did start off using the UT Dallas, antique implementation, I kind of looked at that first, only because it was you know already integrated with VLC quote unquote, but it was also from a much earlier version of VLC that wasn't quite as modular then so it's helpful ish. It was also mostly had a lot of windows API specific stuff, so it wouldn't work as well for compiling with VLC J holin.

M

Both his code and just talking to him over email was really helpful for me to understand, and also just where to get started, because VLC is just really big, even though there is a lot of documentation and stuff kind of confusing to get started. Cisco has an open-source library for SSM amt, so that was also pretty useful.

M

um Theirs was also, though, intended to you, can like set up a relay and a gateway from that, and some of it wasn't fully functional, so I mostly borrowed from it what I needed and then the Concordia implementation I referenced and looked at and was using in my initial research to kind of get familiarized myself with everything going on so next slide.

M

um So some of the major challenges uh I spent a couple weeks partially doing research, I, understand, AMT, but also just the appropriate path, because again, I would keep trying something with a web browser and then find out. Oh, it doesn't have this kind of support. Oh it doesn't do this, so um that was a little difficult. I actually didn't have to duplicate a little bit of the IGMP and IP protocols.

M

In order for AMT to work and just getting my messages to send correctly um AMT wasn't a understanding itself, wasn't a huge challenge, but it was still there as something I had to overcome in the beginning. Right now, also, my mate, where I'm stuck right now, is that I can receive the multicast data, because I send the messages correctly and like Wireshark, shows me receiving the multicast data, but the audio is garbled and there's no video rendering so and I've tried synching it up with the Windows VM I have for the other.

M

The working implementation and the audio lines up is just dropping like packets randomly and the output from VLC is like skipping 30 bytes of garbage, so I'm just stuck right there, because it seems like I'm, not unpacking. The packet correctly VLC seems to be confused by some of the headers I will say it was much easier than expected for me to compile VLC and actually start on writing the module, because I haven't actually contributed open source before I'm still in college.

M

So it's just I haven't done it yet, so that was much easier than I expected. So next slide, so I still have it to do this for the rest of summer. I have some weeks a couple weeks left I do want to add in IgM V to support, because I've been focusing on IGMP v3 I. Don't think it should be a huge leap from what I'm doing now. I need to make sure I have no memory leaks because VLC most of their modules are in and C, so I just need to make sure.

M

There's no memory leaks, I, don't currently have the timers fully working just because of the way I'm receiving data right now and I was mostly mostly focusing on the actual functionality, rather the timers. Just yet I mentioned before VLC looks like it does, have a browser plugin, so I could once provided this module would be like accepted into their access. You can make a browser plugin.

M

That's then easy where you know someone clicks on a website and it's like I want to watch this multicast stream and when it doesn't work natively, it just does AMT through VLC.

M

The end goal is, of course, to successfully stream full video and audio because of there's um a relay already set up at Thomas, Jefferson High School near here so and then later on, submit a patch request, feature request to VLC so that this module will be in there full thing, um yeah. So um next slide I know Oh.

M

Impressions so with a MT again, I didn't actually know what a MT was before I started on this project at first I was like a little confused because it didn't fully make sense to me, but then, like once I like read the RFC and into it more or some of the Articles that Lennie sent me I, definitely think it's really good for solving, as he called it like the chicken and egg problem, where it's like, there's, not that much multicast content or well.

M

There's also like not an audience because there's no content, but there's no content, because there's no audience so I think it's a really useful way to help transition into hey. Let's have more multicast content available for as many people as possible. I did think the RFC 7450 was pretty well written, very thorough. The diagrams were great, so VLC itself, just briefly, their documentation is better than some of the other open source projects that I've looked into before or had to like glance at for school and their chat and forums were okay.

M

Just from my experience, I had some issues that everyone kind of dismissed, right away and I was like okay, but it was really easy to start developing though base on their guides, for it so um I think that's it for impressions. So last slide. I. Think.

M

So um yeah any questions or comments.

B

Cool well, thank you. Natalie you're, a kind of an excellent.

B

A good case study for this working group because amt the spec was essentially written for someone like you, somebody who doesn't necessarily know multicast but can walk in, and and so it is good to hear that you were able to pick up the spec and figure out how to start writing and implementation. So that's actually great feedback for this working group.

J

Greg Shepherd from Cisco. So what are your next steps.

M

Right now, I'm still trying to finish up and actually get the video stream, in addition to the audio, because um it's not fully synched up right now, like I'm, definitely receiving the data. I just don't think I'm processing the packets correctly. So it's kind of a matter of like checking my sockets and checking that VLC is actually getting the data correct. Since again, it kind of jumps through a lot of Hoops. So the most immediate one is fixing that.

M

But then, after that, it's going through the steps to do like a feature request and like code documentation and all that.

J

Excellent. Thank you. Yep.

B

Thanks: Natalie yeah, okay Jake, you are up.

J

Hit the big red button for him: is he and do it that way.

B

Well, we didn't need to edit for Natalie, so I'm assuming interest for.

E

Jake Oh, once we hit the button it's been hit. Yes,.

B

It's you one hit at once. Gotcha.

O

C

There you go hey.

E

O

Though I held my comments, cuz I know I was going but Thank You Natalie, that's exciting work! That's I'm interested in hearing more about your progress, see your end up, work great, so I'm going to talk about our new proposal for asymmetric manifest based integrity. This is meant to provide integrity, guarantees, cryptographic, integrity, guarantees for for multicast instrument. There I think, there's no reason you can't use it for unicast, but it particularly, maybe like AMT, would be a potentially useful case to do that, but you know so you're.

O

Looking for multicast um yeah inter domain multicast, we think that one of the issues it may be. It's not the only issue but I think it's a blocking issue that we're not going to have a real and bone with inter domain multicast that you know it is friendly and hardly available if we can't make good integrity, ultimately, I think- and so this to me we're trying to fill a hole in the what we've got.

O

So you guys know what a multicast but we're trying to solve the main multicast problem and that's one of the points of feedback I want. If I'm talking about that, then I would like to understand it and know that. But but that's the sort of premise behind this yeah next slide.

B

Jake I think, or at least the other folks, on the on the Medeco. I we see you, I don't know if they see the slides. Did the me deco people see the slides.

G

They still okay, good.

O

All right, no revolt, all right, so so our goals here are about. We want a hat. Mostly. What we're trying to solve is live video, that's kind of the main use case we want and we're looking for asymmetric crypto. So we mostly want integrity, because when we're talking about live video, especially to a wide audience, you know symmetric keys, just not like there's a point right um and we basically want to be able to do it it and handle lots of work really in the next slide.

O

O

There we go so let me preface this the next three slides, including this. We got some great feedback yesterday from psych dispatch. Eckhart pointer does pointed us at a really interesting paper that I think might be able to this or help us improve on on the schema that we have. But what I'll describe here skin in the draft now and just take hope that we might be able you a little better. So the basic idea is, if there's an anchor message up on the top left.

O

That is the that that's fetch, eerily usually by a unicast.

O

A security, a caste connection, but we say any secure transfer is fine or you can verify the authenticity and that contains public keys. The public keys can be used to verify a signature. That's part of a manifest the manifest in addition to a signature.

O

Riya has hashes of the data packets that it's authenticated, so this provides a proof that cryptographic proof that he sent her, that the owner of the private key listed in a securely transfer to anchor those the most the data packets that were part of this data string. Okay, so next book next slide.

O

Yeah, so the we extended this, because we noticed that it's gonna have scaling problems so when you're talking about like a 20 megabit flow you're, looking at like 2,000 data packets, a second depending on what kind of security you think you can get away with in your hash thefts, you're looking at like 140 to 100, is per packet at most with with a zero length signature and we're not sure how big the signature has to be.

O

Necessarily if you want to use RSA, if you want to you, know a big RSA, then and you're gonna take a chunk out of that as well. So by making a treaty we can so. This is kind of like a Merkel tree structure. Here. By making a tree, we can have a signature at the root of the tree, which then provides integrity for child, manifests that don't have to look themselves hold a signature, and the child manifests then have the hashes of the of the data packets.

O

So this still provides the very similar kind of integrity guarantees for the data packets themselves through this chain of manifests next slide.

O

So for redundancy, of course we need you can either retransmit the manifests that you want, or you can send the data packet hashes multiple different times, signed by a different tree different times it's. So this is a trade-off of a signature rate versus you know: redundancy versus latency of the history you're trying to secure right.

O

So that- and so this is, these are all configurable as part of the anchor message to decide to determine like what is it that you're going to be sending and and send it in a manner that it matches and as long as you're getting the assigned hashes of data packets, you can verify that all the data packets receiving are the ones who expect or reject Billingsly. Yes, the next level.

O

So one important piece: that's all these things, I've said so far. You can sort of get by Tesla. All right test was a spec from a while back, which provides the way they good is, is a symmetric key, but they don't release the key until after everybody is data packets, and then they send the key note, and then you can tell they is the key, the data packets, that we had the ones that were assigned by the key that just explodes on very signature.

O

So one of the new capabilities that we provide is that the intermediate nodes can provide authentication as well, so you can avoid sending authenticated traffic into a network if you put it at your engines, if that's the intent and though one of the points we noticed was that an attacker in a setup like this, where on truck they've, got a way to send untrusted traffic as a man on the side, so you're adjacent to the path, not necessarily on the path.

O

You can smooth the traffic here and you can, when you have receivers, joined to a to a stream that you're attacking you can make a congestion attack which the IFSP layer might not be a significant amount of traffic. What might still overrun the what their suitors Candis looks like.

O

So the way we do, that is that the intermediate ambi system has to get the anchor and be able to process these these packets and authenticate them before forwarding. So the idea is this is an idea we stole from from the ante relay discovery from a couple years. Back, that's still languishing might be worth looking at again, but the the idea is an SG joint is issued and because we have the source as part of the SGA, as the RPF signal goes back to the to the ambi.

M

Node, that's an intermediate.

O

An intermediate system: it now can do a DNS lookup on the universe, I the reverse source IP and we are I- think it's not in the draft yet. But the plan is to make a the NS type here and to use that to expose the URL for the anchor message and then the anchor message provides all the information necessary to authenticate all the source source. All the traffic from this source is being authenticated through next time.

O

So next steps we of course, are looking to improve protocol. We have the.

O

Our research to look on how to how to integrate the advice we got from sec dispatch so far, the other advice from sec dispatch was. If we want to talk about opening unciekin, we need to go to find people who are interested in this.

O

So it's really hoping that this audience would be universally interested, uh-hum or close to I and I understand some people may not think that inter-domain multicast is one thought, but my opinion, that is the essence of Engram and so I'm, hoping that we can look for ways to move this forward, and you feedback of course, is very welcome. Next slide, it's just more asking for two things and yeah.

O

Right, the DNS thing I haven't seen it a lot I'm interested in hearing. If people do have problems with that idea, this figure we can understand better. Obviously, if this whole idea is it's just not worthwhile, I would like to understand that as well and walk away before I spend too much time on it and outside that yeah I would like to make it just as solid as we can were, depending on the feedback we get. We are probably looking at taking a stab at at an implementation.

O

Sometime, maybe maybe we'll have something forward motion before.

J

Hopefully so Jake Greg Francisco got a question for you, so you mentioned emsac. So is there a bit of a gap analysis to see what m/sec, what ball do they drop that you're trying to pick up here right.

O

So I address that a little bit in the draft. It's the under the heading of comparison with Tesla. The basic problem is that Tesla doesn't provide non-repudiation, you can't, and so the issue is that, if we want to, if we want to let both the receiver perform, authentication and a network element perform authentication, then Tesla can't do it because by the time by the time you release the key either the the intermediate network node has forwarded.

O

Untrusted traffic or the receiver is not going to get a key you're not going to have before they get the key.

J

O

An excellent thanks.

J

J

Any other questions comments on.

B

The last one, the feedback on the DNS thing, I think real a discovery. Amt real a discovery is a long overdue topic. Like you said there was the eighteen t guys came up with a draft using DNS. Actually, maybe fifteen years ago, someone from Sprint came up with a similar way of using DNS and I. Think it's worth reopening, not just DNS, but just generally speaking, how do you pick a relay when there's a bunch of them, some of them being any caste?

B

And do you pick the one you do you pick the one closest to the source? Do you pick the most so I think? That's that's. That's. Definitely something worth looking at I. Think DNS can can play a role in that so right.

O

So that's another point: if there is feedback on the DNS thing, there are at least one that being the anti relay and possibly two possibly another thing. I want to consider using the fourth, but I think the trick is pretty good, we're finding a you know: RPF source related thing from an intermediate node, so yeah. If you.

M

Don't like that idea, then.

O

I would like to know what all right well. Thank you for the slop okay, great thanks, Jake.

B

All right, poreless come on down.

B

So you have two presentations. You have 25 minutes scheduled for 25 minutes, ironically you're right on track. So you have.

J

10 minutes tourist time 25.

B

Okay, actually.

J

B

You have the 30 minutes after that, until 10 stuff, you have until people leave yeah, you have the cookie hours, but my suggestion is: do the deprecated one first and then do the other but yeah? Oh.

E

That's right: let.

R

E

It doesn't work. Let me try. Okay,.

I

A

I

So lost without you, okay, so anybody still remembers ITF 101. So we discussed basically at that point in time the kind of revised new draft that came out of prior drafts, in which way you know, started trying to deprecated ASM completely, and so we wound that down to deprecating ASM in the inter domain space, not but not in the intradomain space.

I

So that was the agreement, hopefully basically with 101 the 0-0 draft of that, eliminating the resistance that we had in terms of trying to get all the way into deprecating ASM all the way- and you know one of the reason, of course, is that ASM intradomain is still widely used right, so there is really no replacement for the best form of ASM by dripping in enterprises. You know to replace you know: 20 thousands s, kimochi state with a few group states, and we can still think about. You know, starting to retire more intradomain I.

I

Think operationally. But you know really. The primary issue from the operational perspective is that it's a really good thing to start deprecating on the service provider side, because that's where, basically, the service providers have a good impact on getting the applications to move to SSM right, they run their IPTV shop. They basically are in much better position to get the SSM adopted, while in the enterprises we've seen that for 20 years, it's really very hard for them to force applications to support SSM if we're lucky.

I

Basically the stuff that happens on the service provider, side SSM wise will trickle in the enterprises as well. I mean these days. What new applications exist, they're, all the ones that are being used in the internet. Now the enterprise isn't really the place where innovation happens. So that's why I think we're going the right order of you know not touching the enterprise's inter-domain right now, but you know even something like internet to. If you remember the discussion we had in before.

I

As being one of those networks, that's trying to get rid of ASM, that's connecting a lot of enterprises, research organizations right, so that would you know already start going. The direction so that is, zero, zero.

I

Yeah next slide, I think yeah right. So then, basically shortly before 101, too late to to bring it up in 101 I, provided you know a bunch of clarifications and additional operational inside, and so that became version or one I got onto the author's list there.

I

With you know, explanations like the one I had on the previous slide, adding text to discuss the MSTP challenges and why there is no MSTP for ipv6, so kind of a bunch of the background providing further evidence of you know how really dead ASM in real inter domain actually already is, and so you know, maybe we can still rephrase that we're really not killing it right. It's basically already dead. Mstp is really used in very few, inter domain deployments already only so yeah. So please read through the 101.

I

Was there anything No, so I think the the key part of that was at the bottom point right: expanding the scope of inter domain, so one of the tricks was what is it actually intra and Inter domain? And you can have different views on it right. One is for the product called site, so app in person will tell you Oh inter-domain.

I

You need to have different PIM domains, but if you come from an operational perspective, it's good enough to say you have multiple administrative entities that are managing around in the network and that definition, if we take it operationally, really allows me to say a single service provider that has you know a million home networks with home gateways together is an inter domain case and avala I've also made what we're saying inter domain be applicable to the most simple case.

I

We really want a SM out of the picture, and that is the service provider offering IPTV services that goes to a million homes and where you definitely don't want to muck around with in sparse mode service provider tried it. It took them in many cases, six years to get the bloody IPTV application vendors to support SSM. In many cases they had to change their vendors to others, but now we're really in round two around three of these IPTV deployments and I would hope they are pretty much as the same everywhere.

I

I

Right so- and there is a lot more additional stuff, so we added text to also suggest to prefer SSM intradomain when possible, right and discuss the application dependencies when and where that's not possible, edit stuff to suggest documenting in some BCP for how to develop SSM applications who really support this move right, because I think there are things that we learned like you know how to get SSM in iraq into electronic program guides for IPTV applications or whatever signaling you have wherever you are signaling a group address.

I

Why don't you signal a source address together with it right? So that would be. You know, added work, we're suggesting to be done here, yeah and then oh to was just a fixes right um nope. So this light so- and this is basically the stuff that I primarily wanted to talk about in in the PIM working group, so the associated candidate work would really be figuring out.

I

How can we get the classification of our standards into shape because right now, older standards that no, but he should use, have a better standard level than what we want them to use like IGMP v2? So that isn't a different different presentation and then also? How can we get completely rid of em SDP right, deprecating, ASM inter-domain almost gets the job done.

I

Unfortunately, MSTP is already more or less retired, as I said into the domain, but it's heavily used in Trudeau main for the MSTP mesh group and it's actually doing a pretty decent job for that right. But if we want to kill it, I think we need to basically make the product call that we actually would want to use for mesh groups, which is RFC 46:10 ramped up with all the operational requirements that we have and that's reliability, debug ability, troubleshooting and control what we've learned in MSD P over the years.

I

So the proposal is there one draft that that we have in PIM to basically get us to the you know: transport over TCP for the reliability flow control between the RPS and the other. One would be yang model that somebody still needs to write or hopefully find a young author, where we can add it to one of the existing yang drafts or updating them. um The ssme application development recommendations that I mentioned in before I'm, not sure what other you know outcome. We want to have on really evolving the standards bases operationally and protocol.

I

Wise of you know our you know core framework, but these seem to me to be the most important pieces that we could think about next letter so conclusions. So the authors think the document is really in good shape right all the strategic concerns for adoption should have been eliminated right, and so, oh yes, so overlooked that. So that's just who was about the process so reminders to the list may be, may be good, that's great, so yeah. So in London.

B

We did a in the meeting and I think it was every you know. Everybody's hands were raised in favour. We took it to the list last week. I believe we still haven't heard from. There hasn't been much replies on the list, so we can't adopt this until folks speak up on the list, so what I would say is so first off, let's just retake a poll and make sure nobody's mind has been changed in the last four months.

B

Whoo-Hoo here who here has read the draft.

B

Okay, who supports adoption of this draft.

H

Just to clarify, can you remind me what deprecation means like what's the effect of that in the idea, Webster's.

I

Defines deprecated, as we I mean at minimum I hope it just represents a you know. A statement from the working group, but yeah I mean what official you know levers. Can we pull yeah.

B

So we don't have a police force. The best we can do is write drafts and BCPs. That say, if you're going to do, inter-domain multicasts, don't do ASM do SSM. So it is direction to the world that when you deploy multicast, so what does that really mean? Because again we don't have a police force. You know jackbooted thugs, that we can send out to knox everywhere to start turning off ms DP and other things or filtering we're telling people to filter. You know everything. That's not 232 /e. What effectively it means is it can help.

B

You know the mbone, which we are mbone d. The mbone does exist, it is the multicast enabled portion of the internet and it is primarily i2. So i2 is probably 90% of the mbone and they continue to maintain an MS DP infrastructure. They continue to support inter domain multicast, and so it would give them the ability to- and this is one of the drivers to turn this off and explain to their members. This is why we're turning this off. You know the the the it is.

B

It is best practices to not do inter domain multicast ASM to do SSM for inter domain. So maybe maybe two comments.

I

On that right, so I think the this target status of the draft that we're asking for is PCP and then I think. The way that you pitch the to customers is a BCP is like a standard not for product called spot for operators.

I

Rights I mean that's I, think the the simple pitch or so, and it's hard to somebody to challenge it because I think that's pretty true right, the second one. Can you send an email about this? You know filtering of 232 I, don't think we had that recommendation, part in it. So it's a good one that no.

B

No I we're not recommending it. Oh you don't recommend.

I

C

B

Remember to in in London that was one of the things that was suggested and it was next okay, so yeah I know I, don't think we want to go that far, it's more of a finger wag than a a jackbooted thugs coming to your network, to turn it off. Okay, but another benefit Mike, is to give clear direction to application providers outs.

B

People outside of this working group, who just think that, oh we'll just you know, join a group that you you need to do, SSM that that here is you know a BCP BCPs are saying you know, there's an RFC that officially says: don't do ASM do SSM, so it helps providers. It helps operators give clear direction to their yeah app app developers, yeah.

I

And so all the folks who race the hand- please don't sleep on the email list for the next few months say yes, when the yeah? No don't.

B

It's not for the next few months. If you want to send an email right now you can so so you know when we took that informal poll. It was you know five, oh, so that was I. Believe five people raised their hand to support it. Anybody against supporting this, this draft, who doesn't want to see this draft adopted?

B

Okay, so it's five zero. Please folks send an email to the mailing list that says: I, support or I do not support, but we need to here. We can't adopt this unless we get again unless we get enough replies that.

I

The promise just the other half of the working group is co-authors, so yeah yeah all right. Next.

B

Other questions about for deprecated ASM moving on right so.

I

This is just for informational purposes. What we have the time, so we can talk about it. You have somewhere between.

B

11 and 61 minutes, depending on how much by the new yeah.

I

No, that's fine! So let's go through it next slide. Okay, so problem statement. So if you are trying to adopt a product called from a host to a network to signal, multicast would be the highest. You know internet.

I

This is a Q&A session.

I

Right, so that's that I think the best pitch I can get about the problem statement. So so in so v2 are mtv2 and IGMP.

B

V3 never made it to another net standard no proposed standard Wow do.

I

You know why I have no idea, probably because we've all been been asleep at the wheel and never cared about processes. Have you been at that RFC I on Monday evening kind of process is really not a nice thing to think about, but you know after decade. Maybe you know just investing a little bit in this process to basically make sure at least that stuff isn't becoming a reason why people do the wrong thing. I.

L

Think I think this. It's just like a natural process or kind of like how ITF works, and you know I hear small has been there for a long time if they came kind of mature and sure we make it a standard, of course, in the later versions they have been around for shorter time. So just I'm.

I

L

I

Years instead of 30 Sharia is.

L

What I would like to yeah maybe see in the ITF in general, you know what happens when you come up with newer versions like your protocols. How do you deprecated the thing maybe making.

I

It historically, this.

L

Is the problem right so.

I

There's basically, this poor app in working group, chair or area director that I'm going to ask these questions in in some session in an hour or so right, yeah.

L

So this my thought would be person. One should probably become historic toward a prohibit, but double talk to these and others to find out and- and of course, we need to look at. Probably you know either worse than tours and see. If you want to make something a standard you generally need to go through, like implementation report, make sure well.

I

Are you going to say something different now it's an individual mbone contributor than what you would be saying for the same question in our.

I

Right so yeah, let me quickly go through the slide, because I also had talked with operators. You know, after this, the deprecated stuff and about you know we want ASM and SSM right. So so the this stuff is related right. So so the operators think they do not need to go through IGMP version 3 or ml db2, as long as they do not need SSM.

I

That is, you know, to some degree true right, um so it would be good enough to stay with IGMP v2 and MLD v1, but they also get other benefits from ml db2, as are it from IG MPV, ok, typos from IGMP v3 and mo d, bad typo, ok right so one, for example, is the explicit tracking, which is one of these other issues. Where I think. We also need to point out that you know these latest versions to provide short live latency, which an IPTV is extremely crucial and then run all charter.

I

You know March our brains, what beside SSM and explicit tracking, where the other benefits that we get out of the latest version of the protocols and that's basically something we could write down and into an informational, a draft in Ambon d, basically to go along with these standard document status, changes.

L

Thing, I think, the main problem that I see all over the places that people think are your peers and thieves as a same homeroom right exactly.

I

Yes, yes, so if we can have an RFC with 20 type, bold font, that would certainly may be the best way to emphasize that that is not the question. I can.

J

Your bullet there's a little misleading as well. Craig Cisco says wrong. Fully. Backward-Compatible can mix two and three well, you can do as a SM without I mean v3 supports ASM. It's not like. You have to fall back to 2 or 1 to get ASM functionality. V3 supports ASM yeah, of course the well it's just it's confusing backward compatible can mix v2, v3, routers queries and arbitrarily, like yeah. You can, which you don't need to. You can do ASM with v3 I'm, just right.

I

J

That the customer, through.

I

It there are going to help. You know. You know, of course, of course I'm here, for you Torres. Just thank you. Thank you so much yeah, so yeah the midpoint.

I

Of course, you can perfectly go through to IGMP v3 and ml DB, even if you'd have nothing on board with ASL with SSM, because your intro domain, you spider film, whatever good stuff you want to do in your domain, don't stick to the old protocol version right and then also any mix should work, and now we've seen basically the problems in the past, and so I mean that's operational. Experience that we, you know, and an embodied raft may want to mention.

I

I- would think that you know my latest problems with IGMP v3 went away in something like 2010 when the last snooping switches started to support it right so I mean if we have some timeline to to really give more confidence to operators. That would certainly be not wasted right so and then, of course also when we are writing something operationally down, then an MD document would be the place to state.

I

You know here are all the the changes we've been doing and deprecating ASM inter-domain, but we really want to have you know a full internet standard for intra-domain ASM and that's basically around you know the the other draft I mentioned right, so the deprecation in the domain, and then these these recommended these discussions right. So intradomain is fine by their premise. Preferred majority of customers I would think the majority of multicast deployments.

I

Overall, if you just count, customers would still be enterprises that are using PIM, sparse mode within the domain and then obviously, all these things like there's a good range of application that could do SSM, but we're not forcing you now, but you know whenever there is an opportunity right so I think that that statement should be made, because that was exactly why operator called me and basically said he was, you know, afraid about.

I

You know the work that we're doing interdomain really touching whatever he is doing, and then we got into all the IGMP weekly discussion that stuff. So how do we make the you know the majority of what we're still making money from unless and operate a vendor is is just selling to service providers right. But if I look at these typical service vendors that sell both enterprises and service providers, we don't want to really make the the enterprise's feel scared, and so that's this part next slide right.

I

So now the the funny protocol list right so RFC, one one, one two, and so this is actually the only interesting document. Everything else I think that we have on the list is just changing the status right, so the bad ones downgrade the good ones upgrade right, but RFC one one. One two is two things in one on once: it is the actual specification of the ASM service model for which we actually never built an ipv6 document.

I

And, of course, as we said, we don't want to duplicate the whole ASM service model right, even if the last step would be at some point in time, say: intradomain pins, our spot is not good, but intradomain advisor pin is I, think always going to be excellent. For you know the applications it's built for right, so you know that's. This is the biggest challenge here. My proposal would really be to ref this and basically come up.

I

You know just strip out all the IgM p1 stuff, add ipv4 ipv6 in in the rest, and that's going to be the new normative full internet standard for the ASM service model. My pet would be my proposal. Hopefully that's that's something feasible, because this is something the holy grail of multicast right, so upgrading RFC saw would not would try that we don't touch any of the actual content they're. Just you know v4 v6 and get rid of all the IGMP v1 stuff.

I

So then our yeah IGMP v2 right down great historic, I, don't really know the process, I think the the historic process. Hopefully there's going to be an ad and in working group, and we can discuss that next slide.

I

Yeah IGMP v3 needs to be upgraded, I think to full standard ml d v1 downgrade. Oh, there is a typo again sorry. This was all done quickly, yeah, so ml d, v2, of course, upgrade to full standard. Same argument is IGMP v3 yeah, that's V, yeah.

J

I

No I'm just doing the current Kirsten, so this is this.

C

Is the current? This is just.

I

The current yeah.

C

I

Yeah so I mean the upgrade and downgrade for all. These documents is pretty obvious right, so the old ones downgrade the better ones upgrade so 4604, that's SSM! That's the service model for before and v6 upgrade. Then we've got 35 90 we'll have to check. If that is actually you know well adopted. So that seems to have been an addendum to I, Jim, pv3 or ml db2. I. Think so we'll need to check that in more detail.

I

If that should become part of you know the Catechism in full standard, then we've got the most interesting one. Just you know from the status and that's the lightweight IGMP v3 ml d v2, which is kind of removing the nasty bits from IGMP v2, IGMP, v3 and ml db2, which is the exclude sources list which we never really used. So I have no experience about deployment stuff.

I

If but I would fear that that that's still far away from really being full standard right, that's that should probably stay proposed and we may want to continue to suggest to people. You know to use it or not. At least it's part of the investigation group yeah.

L

Stick here so yes, the interesting thing is when you want to make something: an Internet standard, you actually look at the spec and you decide which parts have like white implementation and interoperability, and all of that you actually take stuff out of the spec that hasn't had it. That is not nature enough. In general, like we did, I picked it with him, with between I, actually took some stuff out of him off its border border.

I

L

Saying we need.

J

I

The document actually when we want to upgrade to.

J

Full, what I think what you're getting at is.

I

J

May evolved it look like fifty seven ninety when it goes yes,.

L

Divots and it could be that we actually end up with being that- is that out the way, but there are some deployments, so maybe we will find if there is insufficient well.

I

I have no strong opinions right, I'm, happy if we don't have to go through the trouble of doing this stuff. Just if people stand up and feel strongly that this stuff has enough stuff, then I would obviously agree that we, you know, do the job on that as well. I'm, more interested in what you were saying with respect to what's the process, I was hoping looking at all the documents that seemed to have been updated from proposed to full standard.

I

After a while, but now you're saying we may need to erect the documents and throw stuff away so.

L

Issues to be fixed, probably own implementation report- may.

H

L

Rfc, like they did with me, wrote a sparse mode right.

N

L

Is one RFC that is like an implementation report and then, after that they did like it took some stuff out of 46 so long to make yeah.

I

But there was a young junkyard. There was a 20-year junkyard right, I feel a lot better about IGMP and MLD. For that matter. Yeah.

L

So that makes the job easier. That's good,.

Q

Jeffrey from juniper, this is related to sticks, comments related to the exfoliant sources. If we upgrade the original itv3 ml to be two to four standard, then we neither decide if we want to remove the excluding sources function and I'm. I was actually not aware of a fifty seven. Ninety, if the that I with the RGV 3ml TV, is without the exclusion sources, then we might as just consider upgrading that 57 92 for standard and.

I

That would be an interesting question right if I just take a GMP, v3m, LDV and basically try to change the text. To say that you know the exclude list must be empty right. What difference would that be over 50 790, because I've 50 790 rewrote everything right, and so that's going to be hard comparison. Maybe that's that's one of the tough jobs to do, because I would agree that if we need to arrest us that hasn't been used, it would be exactly that you know exclude list having anything more than an empty list right.

I

So if that's the thing we could do to basically say: oh, we don't need 50 790 we're just removing the exclude list, but there may be other product called simplifications that 57 90, so we'll need to check the authors list and figure out if any one of them is willing to to help with the effort, because you know when this stuff happened. I was on the position of you know. Having worked in the vendor that had spent a humongous amount of time to get the full versions of the protocol is implemented.

I

So there was little interest from our side to basically fix this, so now, basically, with a lot of time gone and we're trying to do the best for the IDF, let's see how we can deal with it now.

B

You've revealed way more than you should have, but that's okay, we did the same thing with a.

C

Lightweight version of.

B

It before it was a lightweight version of it, and we just did yeah.

C

We spent an enormous amount.

B

Of effort doing the whole thing yeah.

Q

E

Question is about v1 if we upgrade.

Q

The v3 to full standard do we have to downgrade the v12 historic? Can we just leave it on? Why.

B

Not downgrade v1 and v2 days, I think.

Q

Torres said they had a concern that if we downgrade of historical, then we lose the definition for the IAS yeah.

I

But remember RFC, 1, 1, 1 2 doesn't even talk about ipv6. We have no normative document for the ASM service model in six, so I think it would be definitely good to to ref. You know, then you know split one one. One two, you know just extract from one one. One two, the service model, which is the 50%, then comes the I gene, p1 spec, so we'll just take the first part at v6 to it. Something like that. Well,.

Q

It's so appealing, even we didn't do any other things at all. We just don't have.

I

We don't have a standard full standard definition of the ASM service model for v6. So if you want to do, ASM do before yeah sure I mean some people might like that. But again, if I want to defend, you know the poor enterprises that can perfectly well do ipv6 by derp in ASM. What's the reference.

B

Or maybe there's no one doing the six multicast in the enterprise.

I

No I, don't think that's true I think you know the enterprises as much as anybody else are moving to v6 on the time line that they can afford right in terms of it. The vendor I'm.

L

Running yeah, just a just quick comment to Jeffrey, so we don't have to touch anything else in order to move that supposed and ago internet standard, it's up to us or I want to deprecate or whatever, versatile one. That's a different discussion is what sorry do we want to touch person, one like deprecated or whatever? That's a separate discussion. There's nothing automatic! It can make new one without touch.

R

Montgome, not Cisco Systems I just wanted to add one more document to your list, which is igmp, snooping, I, think even that RFC is still international. So maybe it will be good to make it as a standard so that the.

I

Information on that one wasn't because it's not widely deployed or so, but because it's doing things implicitly, it's not changing anything on the wire and they're. Basically, these ideas, policies that are saying show me differences on the wire. Then we can make it standard now we're actually having the same discussion with the explicit trekking where the changing on the wire is basically that traffic flows faster or you know, stops faster and igmp. Snooping obviously does the same thing, but you know I jumpy snooping has this of a layer to solution.

I

That is, you know, so that's a totally different discussion. We can have that again with the area director and see if something has evolved, but the informational was because of that sure.

J

Here the questions, it's good discussion, thanks for listen, Jason, work.

F

I

Oh, that that will repeat this because it was so cool an hour.

J

Who's got the blues blue sheets. Everyone signed the blue sheet anyway,.

B

Yeah, so we PIM will be back in this room and it starts in 20 minutes. So if you'd like you can stay here, no one will be kicking you out, but come on back in 20 minutes for PIM other than that and Bondi is adjourned. We'll see you in Bangkok.

L

So PIM PIM starts in 12 minutes.