Internet Engineering Task Force IETF 114, 28 Jul 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF114-GROW-20220728-2130

Description

GROW meeting session at IETF114
2022/07/28 2130

https://datatracker.ietf.org/meeting/114/proceedings/

A

Oh not too bad.

B

C

Are you starting the chair, slides? Oh, you already have some up.

D

How do you control it.

C

Welcome everybody welcome in the city of brotherly love, the.

E

C

That is always sunny philadelphia. Also a warm welcome to our remote participants. We appreciate you uh taking the time out of your day to attend this session. Even if this time zone perhaps is not optimal for you.

C

Today's working group session is the grow session. This is the global routing operations working group who concern themselves with the stability of the internet routing system, as deployed across this planet, a phenomenal achievement. uh Despite our best efforts now uh co-chair, would you be so kind to jump to the agenda.

E

C

A note while applying it applies. This is the fine print that outline outlines your obligations as itf participants of special note is a desire for everybody to interact with each other in a respectful manner. um I also want to highlight that it's mandatory to wear masks during this session in this room.

C

I see at least one person not wearing a mask, so I would kindly ask you to put on a mask uh much appreciated for for your collaboration and let's hope that, maybe in a few years mass or not a requirement anymore.

C

The note will can be downloaded if you want to read the unreadable. Letters.

C

There's various uh links that apply to this working group session and the working group in general. um Again, the materials are on the data tracker. So if you want to read all of this in more detail, please download the materials and follow all these links to resources.

C

Our agenda for today uh we need to dive through some administrative. How do you even pronounce that word administrivia all right.

C

The blue sheets used to be a paper device that was passed around the room, but nowadays it's a qr code that you can scan on this screen.

C

If you whip up your phone scan the qr code and then log in with your data tracker uri, the system will register your presence in this uh meeting, uh which uh has uh purposes in in mostly the legal realm.

C

So please scan the qr code to declare your presence in this working group session. We're going to need a minutes, taker and jabber scribe. Do we need to appoint a volunteer.

C

Richard patterson, would you be so gracious to honor us with minute minute, taking.

C

And warren kumari volunteers to do jabber.

E

You can either do it in the you can either do it in the watchmaker, call it or just send a file, an email to later. Thank you.

C

Any plain text documents is fine, excellent um agenda bashing. We have two items today, uh actually three, because we also should discuss the state of uh various working group documents.

C

But after that, we'll move on to an item uh called the well-known bgp, anycast community or well-known any kind of php community and then more work on bmp and yang.

C

So is there any things that people would like to add to the agenda?

C

Maybe ben madison, with sieber.

D

Ben madison work online. uh I'm not sure that the thing that you are referring to is quite ready to be discussed here, given that there are authors who I probably need to discuss it with. First.

F

C

Then we'll see more of this at itf 115, hopefully um document status. Do you want to cover that chris nope nope from the top of my head, uh the bmp tlv ebit draft, was approved for working group adoption some time ago, but unfortunately I uh it fell off my radar and I uh approved it as recently as this week. So that's now in the system, as working group document.

C

Another document that got approved today for working group adoption is the work on nrtm version 4, the near real-time monitoring protocol. A new renamed document has not yet been uploaded, but be on the lookout for that one.

C

And I see I have to do some work. The grow bmp tov document needs a write-up from the chairs and I will do that in the next few days and I'm hoping paolo will remind me to actually do it in the next few days. Thank you, paulo okay. With that I would say. Let's move on to a remote presentation by maximilian wilhelm, the anycast community.

C

I will now transfer presentation rights to.

C

All the slots for the requested.

C

Oh, you have to pass it.

C

So maximiliam, uh what we're now doing in the system is uh making you the presenter and that should allow you to click through the slides.

C

And we can see you on the screen. We see you waving, there is no audio yet.

C

All right, let's try again.

G

Okay, is that me, or is it you.

C

It was us my apologies, uh we can hear you loud and clear. That's good.

C

A

Able to see me.

G

Again, yes, and I accept what what do I do, um I guess this one.

F

So I see the slides.

G

And you see them too, I guess right. Yep.

C

Take it away maximilian.

G

Cool, thank you good evening to philadelphia. So I put it forward or we freddie is also present in the audience, um put a draft forward to define a well-known bgp community to denate the dote prefixes are used for any cast and we want to say what we saw or what is the reason behind it.

G

So short motivation who am I networker open source hacker work by work for cloudflare, who also sponsoring this? You can find me on twitter or reach out by mail. So why did we do that so assume?

G

A swiss isp, who has international network intercontinental network and, for example, wants to transform transport packets to the us directly so would like to prefer the own path where own equipment over the pond, but would also like to route any cast prefixes locally because of the nature of any cast for a lot of providers having different pops around the world. Usually the hot potato path would be best and given many providers also use dns, anycast and then point customers or clients to the preferred servers. This might influence some user latency and experience in general.

G

So then we came up with the idea. Okay, if we could denote prefixes, I used for anycast. That would help people to do more, better informed, routing decisions. So the idea is to specify a well-known bgp community to do that.

G

So not everyone um or every provider has to define a different one and t gets messy.

G

That's basically the idea and the proposed solution is, as I said, define a well-known pgp community.

G

We will ask operators who want to denote their prefixes are used for anycast to either specify it on all places or in none.

G

So it does not make total sense to only specify it in a few places, um and the idea is if this is specified, we ask people to use hot potato routing for it, but that's in everyone's best judgment, as was pointed out on the mailing list. This might have influences on route server operators which may introduce an option to filter for any cast prefixes to take that into consideration when doing te, and this also what also brought up on the mailing list.

G

This might influence your network when doing route reflection. Optimal route reflection was brought into the mix as well as at paths. um I say we led it to operators to define the best path forward there. I think it would not be good to prescribe a solution and that's basically the idea. The draft has been put forward.

G

It's in version 01, it's all out there on the list and we are open to your feedback. That's basically it thank you.

C

And with that, we open up the floor for uh discussion and I hope there will be some some uh sharing of perspectives on this topic, because.

C

The implications must be thoroughly understood. I see somebody at the mic.

H

Hey keir patel arcus, nice draft quick question: do you think it makes sense to use large community and append ess also to talk about what anycast uh community, if any you're doing or you rather prefer this to be on a generic community.

G

Just broke, okay audio seems to be back. I think I got most of the question we thought about using large communities, but I guess people in this audience can't answer the question. If it's a good idea to use them today better, I was afraid that there still might be systems out there which do not implement large communities.

G

I don't know about that. That's why the idea was to go for classical communities, and I believe part of the question was to imp to entail the as which is sending this into the community, which somehow would defeat the point of only having one community to filter against.

D

Ben madison work online um thanks for the draft, I think it's I I think it's an interesting problem space. um I think that there's a, I think, there's a kind of a tension between the engineering problem, you're trying to solve and the social reality that it exists.

D

In um speaking, for you know, for myself and the network that I operate, I wouldn't allow any third-party network that to send me hints that alters my local routing policy without a substantial degree of trust in the person who is sending it, and it seems to me that, because that mechanism requires some degree of social interaction, whether it's kind of like remotely over email or forums like this or nogs or peering forums or whatever it kind.

D

The fact that that trust needs to be established anyway means that there's plenty of opportunity to to exchange operator specific communities during those conversations and those relationships anyway. So I think that you end up kind of losing the benefits of having a single, well-known community with universal semantics, because you've got to do all of that social networking work to make it useful anyway to make it trustworthy.

D

um And so I wonder how useful this end up ends up being in practice.

G

Fair point: um the security side of this or the aspect has been brought up on the mailing list, also um I'd say in combination with rpki. It should be okay to filter on community and is our community and prefix list potentially.

G

But I still see the point, so you should think about what you are doing when you implement that.

D

I would like to live in a world in which rpk could help with this, but that's not a world that we live in currently um trying to make secu trying to secure communities is a hard problem.

I

uh Hi there, tom hill from bt, um the all or nothing solution as a proposal is interesting, even as a should, rather than a must. I can see that going very wrong on the basis that either there are transient issues or perhaps even folks who aren't paying enough attention. Maybe so is there a way in which you envision? We may track hygiene of prefixes announced with this particular community, and you know, make make make the owners of the prefix aware that they've perhaps not done this in a all or nothing fashion.

I

Otherwise, we we, I feel, like we stand to be in a situation where that's very easily going to become quite a mess.

I

If past experience is anything to go by.

G

I see your point given that I got shamed on twitter a while back for doing a route leak. um I know these solutions exist and can be built, so I guess going on route, information, services, etc. This should be possible.

I

So you suggest that the current routing leak, detection services or apparatus be extended to point out when the anycast community is not in an all or nothing fashion advertised in all or nothing fashion.

G

I guess suggest would be a strong word for my taste, but what I guess would be a good idea for operators who want to go down this path and announce their prefixes. With this. We could ask them to monitor it, given route views, risks or etc. Does that make sense.

I

G

Makes sense to.

I

Ask them yes, definitely.

G

You don't seem convinced.

J

Jeff, I'm mostly here to remind this working group and my esteemed co-chair of idr. There is no such thing as a well-known large community. Stop trying to do that.

H

Well, arc is here again one for one more time I that was precisely what I was trying to indicate that while there is not one, you could probably define one and get away. I I have a feeling that you do want to tell your neighbor asses, even when you are sending out the well-known communities to tag at least and say, which is that these are propagating from or originating from, and you are appending it and so forth, and so on.

G

But don't we know that by looking at the as path.

C

To to interject here from the as path you don't know which of the asm's in the as path attached, the well-known 1997 community, but there is no large communities registry for this type of functionality.

C

So it would be good to have a discussion on the mailing list, whether a uh extended community that has room for the sender, asn or setter of the community.

C

But uh opinions differ on on whether extended communities are uh useful in this type of work environment. So that's why I would discuss it on the mailing list to to see what comes out of the woodworks um rudiger you're actually last in line. uh First, we're going to go to linda dunbar, linda.

K

uh Yes, um maybe this question is for the group, so I'm wondering like since then anik has used so widely. Can any cast being treated like a multicast like all the ports belong to that anycast register for that, so that the network know how to forward instead of putting to the extended community just the question.

C

I I'm sorry I didn't think I I heard your question. Would you mind typing down the question in the chat to overcome potential audio issues.

K

C

K

C

uh In the meantime, we'll jump to david's dominant.

F

As an implementer with a little clue, um I have a question about the trust relationship brought up earlier.

F

Isn't this really two separate things, one being the advertisement in a permissive way and informative way about tagging, something as an anycast prefix versus having an expectation of something happening due to that, like? I understand that the latter is something that will rely on relationships like that, but I don't need a relationship to make use of this to just flag things that I'm advertising.

D

Sorry, um yeah, that's that's totally true um and from reading the draft it doesn't sound like there was an expectation that this would be useful just as an fyi, but if it is useful, just as an fyi, then sure there's that then it's then it's absolutely fine. um My my concern was that, if there's an expectation that people will actually you know blindly or semi-blindly act on this in the routine policies, I think that's a mistaken assumption.

G

C

Next up, mr falk, thank you.

A

uh Just comment: yes, large community space has not yet assigned any parts of the of the name space for special purposes, but the space is large and interesting. Things could be done and, in my not so humble opinion, probably should the yeah well, okay. It is quite clear that uh the attachment uh of um communities can be manipulated by any as on the path, um no question about that.

A

But uh my guess is that uh the assumption for any caste tagging would be that every time uh exactly the originating, as would would do the tagging um kind of that some intermediate, as uh has the knowledge that yes well, okay, this transiting announcement is anycast and not getting. The signal from the first place looks a little bit strange.

C

So one second: can the people using the floor microphone not only speak into the microphone but also speak a little bit louder, because remote participants are telling us that the signal is a bit low and if there meet echo people on the link.

B

Chris is wandering around and poking at stuff, so I'll carry on talking until how's that far okay, perfect is.

C

This better rob austin.

B

Hello rob austin. How is this.

B

Excellent, so warren, so warren kumari with no hats, so I mean networks kind of generally decide if they want to do hot potato or not, and that's the decision that they make. There's nothing here that actually stops me from just applying this to any old route that I feel like. So, if I send you 500 routes, I can tag 499 of them with this, or I could tag one and it seems like I can choose whatever routing policy I would like with this.

B

If, instead of this being called sort of the any cost bgp community, it was like, I wish you to use a different routing policy community. It feels like that might be a better sales pitch right, like this isn't specifically for any cost. This is, please use your non-standard policy for this prefix.

G

hmm The intention was to use it for any car, so people can but make better informed decisions on what approach to take.

D

Ben madison, again um just on the subject of whether it's helpful for there to be a um for the for the sort, the the attack, the the as that's, attaching the community to the prefix, which is probably the origin being encoded in the community. I think that that kind of assumes that this is useful more than one hop away and my already limited inclination to change routing policy based on this kind of a community would fall off an exponential cliff with every additional as hop in the path.

D

So I think that if this is useful at all, I'd be amazed if it's useful more than a single hop away. Personally, so I think that rather loses its utility again, if it's not just you being used for informational purposes, if it's purely informational, then sure that's additional signal and signal is good.

C

Thank you next up a remote participant, freddie kunsler. If I'm pronouncing your last name correctly,.

L

Well, sort of it's actually clear, but that's okay, uh good afternoon good evening good morning, um I'm one of the uh proposal and actually um maximilian wrote everything, but I could probably give and I the idea why it came up, we being the swiss isp. Maximilian was talking about and some years back we were selling transit.

L

I think it was in frankfurt to a network which was selling transits uh somewhere to network somewhere in the middle east. I think it was iran and they were transiting quite a lot of um announcing quite a lot of prefixes, and one was actually one of these dns anycast servers. I've got which one.

L

But uh what happened then, when you, when you sell transit, you usually uh do them higher local craft and then peering or or um for upstream so our users, our uh broadband customers in switzerland, they were actually then using the uh um dns anycast server in the middle east, somewhere with high latency, etc, etc. If we would have known that that specific prefix is actually in any cost prefix, we would have filtered that away from the transit, and um so we we learned it the hard way.

L

But if uh those any cost operators would actually denote their any cost prefixes it just would make my life a lot easier and the quality of of the service we provide to the end customer would increase. Of course it could be abused, but usually um the any card operators they know what they're doing so. We could expect some more clue of any cast operators than sort of normal operators, and that is basically why I ask everyone to support this proposal.

E

Okay, any more questions or comments.

E

I think there's still some discussion. Have you had about this on the list? Oh go. Are you getting up to ask? No, no okay! I think the next presenter is trying to move to the to the mic. Okay, so I think.

C

um I guess a valid question at this point in time is: do we want to work on the anycast community as a working group documents.

E

I think that's a question for the list.

C

That's what I meant um and a final comment. uh I think the current draft doesn't really define what any cast means in this context. So it would be good to add a paragraph clarifying uh a term that, amongst network operators is uh pretty well understood, but for the purpose of the itf internet draft, it would be helpful to to explain what anycast means in this context.

C

um Okay, and with that, we jump to bmp and yang.

G

C

Thank you so much maximilian.

M

So hello, I'm camilo cardona for ntt.

E

He has actually pointed kind of directly to your mouth yeah. That's right! You can do that just hold it.

M

Okay, so camilo cardona from ntt and I'm going to give update into the bmp gen module draft. So next, please.

M

So we added a few things to the connection to the station.

M

One of them was, um we specifically had a choice now to connect to the station, be active or passive, so I do not know how many people use passive in real life, but but it's included in the bmp specifications. So we added that it does add to the complexity of the model, but it is what it is. We are at maximum segment size, keep a lives and session security session security. We basically copy it from the bgp draft model.

M

Jeff pointed us out uh some hits, so we will probably take a look at that again and see if it fits on the vmp one and how to do it. I keep allies, we copied it from the tcpm module too. I don't know if it's going to fit perfectly so we'll take a look at that again and there it is so. We also added network instance to the bmp sources.

M

So jeff jeff always points out these things, so he pointed out that we were missing that and now we added them and I'll talk about that in a second again in the draft per se, we basically change a few sentences and add an example. uh We probably need to add more examples in the future. I I really like examples in when, when you're reading models, but that's for the future okay, so I wanted to explain this again, um so it would be nice.

M

I'm awesome that we could get all the information in the world from from these routers, but it is bad for the wire. It is bad for the collection and for analyzing. We don't have process for that. So the idea that we have on this draft is that you as an operator, can specify clearly what information you want and that we do it flexibly, but also with no ambiguity, and I mean in a precise way, so right now we have four sorry four areas that we we use to specify the data.

M

One of them is network instances that we didn't have previously. So the idea here is that you specify which network is from which network instance you want to get data from, and we added an uh an identity called vmpn and types on the idti that allows you to specify that you want the data from all network instances remotes well. The typical remotes from bmp has been in pre and post has been out, pre and post and local rip. Here there are no shortcuts.

M

You have to be explicit on what you want the same for address families. We take the adverse families values from the bgp um jan module, so we'll see how the two interplay but we're taking it from there annum peers, which is um well.

M

This one is a bit more complicated, uh so you can specify, of course, prc individually when you're based on their ip address and using the bgp module, but you can also specify the type of peers that you want your data from and we are, we are doing it right now with the vtp module. We might change that later. So you can say, for instance, give me all the data from our external peers, which is something that we see very often that you don't want to specify.

M

You don't want to get internal peers, you want their external, but you want to take them one by one. So we added that, uh of course, if you want all peers- and we also have an identity for that, we change that from the previous draft and jeff mentioned that, oh okay, um next one please so here's a brief example of how you can you configure these bmp sources? So, for instance, here we want the data from the global network instance as repeating pre from the ipv4 address family, and we want that for all external peers.

M

So in red is those points I apologize for the xml. This is what it is. Okay, so some gaps we have we haven't. We haven't introduced initial initiation delay and back of times, which was um an observation from from team evans. uh We just haven't done it, but we are not. We haven't forgotten it yet, and I'd first like to thank jimmy chen, I'm sorry, I mispronounced his name.

M

We had a very nice discussion about how to uh about all of this that we just discussed about bmp sources and being able to filter the information, the data they really want to send to the collector.

M

He actually went a bit further and he he told us that he actually needed to filter prefixes per prefixes and he actually told us we could use a specific model that allows you to basically specify a a simple router routing policy for that. So we we still we're still processing that, but it was an amazing conversation and but right now we can just thank him.

M

um We better. We were a bit ambiguous last time uh we had some discussion in the mailing list and it was not clear so now I mean we'll just like to make it clear. Is this good for work working group adoption?

M

We can take it to the mailing list, but we can also discuss it here a bit and that's it any more feedback. Please. Let me know.

C

What tool chain do you use to compile the yang.

M

To compile it or to validate it or to validate it, but by yank by yank and some cisco tool that, for the example for validating the example, I use some cisco tool. That is free. I don't remember the name sdnss stuff, yeah.

C

I'm asking because I believe this would be the first young work in the grow working group and they're, at least speaking for myself is a bit of a lack of familiarity with the various tools to interact with young models in internet drafts.

C

Are there any questions from the working group.

D

um Ben madison, thank you for doing this. I think this is really useful. um I think this is one of those areas where interacting via a mailing list or providing useful comments in a meeting like this is is, is difficult.

D

um Is this work available somewhere like github or a similar service, and, if so, is that all kind of linked together from data tracker, so that it's easy to find, because I think that's going to be where useful contributions happen.

M

Yeah, we don't have it, uh we have the public, but we will do it and and of course there as us, other itf uh models have been discussed. We can, we can make them, we can make them public. Yet. Thank.

E

B

M

B

More of a comment than a than a question: there is also the ietf young doctors who are available to help with sort of consulting and things like that on this. So if the working group has wants to sort of get more experience- or you know understand how it works, I think they'd be happy to help.

I

Hello, tom hill, um I believe that there has been a suggestion within netmod to collate, um to try and group together and bring together all of the distributed young models that are developed in different working groups into the same netmod working group. Github repository it's worth having a chat with um the chairs of netmod slash the ad for netmod rob wilton about that.

G

That makes a lot of sense.

M

And we will be also very happy to do it after it gets working group adoption, because I guess that officializes the whole thing right um more than just make it like personal contribution.

C

All right, we'll uh launch a mill thread to discuss working group adoption of this work and then assess consensus on how to proceed.

C

um I see the queue is empty, both virtually and physically, so I think you are done. Thank you for your presentation. Much appreciative.

C

And I think, with that we can wrap up this ietf meetings grow session. We are giving you back 20 minutes of your life and I hope to see all of you at ietf 115 in london.

C

London is on a different continent than the londons that are available.

C

All right, thank you all for attending this meeting and um until next time have a good day.

D