Internet Engineering Task Force 113, 24 Mar 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF113-LSVR-20220324-0900

Description

LSVR meeting session at IETF113
2022/03/24 0900

https://datatracker.ietf.org/meeting/113/proceedings/

A

B

Alessandro good morning, everyone.

C

Hey good morning.

B

Comment, you had the best slides at the right routing area, open meeting winter best presentation the longest one for sure.

D

Yeah, I think that scion guy after he saw yours, he was gonna hightail it back to zurich.

E

Hey good morning guys, can you hear my white uh voice? Do you hear me right.

C

Yes, yes, yes, works.

C

Let's give it one more minute and then let's get started.

C

C

Okay, let's get started so we are like one minute, past 10, so this is one minute past 10 o'clock. So let's get start with our session here. So we have a two hour slot and the agenda is, you know, reasonably lightly populated, so I think we have enough time for q a if that actually is necessary and yeah. So, as you can see also, you know there's like nobody sitting in the front due to unforeseen circumstances.

C

uh It is, unfortunately, you know the way it is right now. uh Luckily, you know beyond the initial hiccup of meat eco, you know having a small issue that actually has been fixed now also, so I think you know everything is ready to to to go as such, so welcome all to link state vector routing working uh first, some administrivia. So this is the noteworld. You probably have seen it already right. I know by now- and it just says, like you know, that.

F

C

If you're participating into the itf and whatever you say as wondrous like an ipr relationship here, you have to take that to be into account now. In addition to that, uh there is also some anti-harassment procedures and code of conduct so be friendly, be kind, be nice and don't make communication. You know discussions like personal, but keep it on topic uh on a technical level.

C

Also for meeting tips- and this is especially you know important now- maybe not so much because we only have like three people in the room.

C

I believe so, if you wanna, you know, uh ask a question: you know, use the tools and then you know we will be able to grant you access to the microphone through the remote access tools via from from the chairs, if you're local just hit the little thingy here, you know what you can see on the right hand, side, and then you will be able to get into the queue through that channel.

C

So we also have a few interesting links. As you can see, uh you probably have seen them already, and you know these are just reference in here. I'm gonna go on further.

C

So for the rest, uh so for the javascript we have a jabber, you know tool so because mid echo is directly integrated with it uh for minutes. uh Victor is going to take care of that one and then let's go towards the agenda of what we have so as I mentioned, so we have a reasonably light agenda and we have like a long time schedule planned. You know ahead of us. We have three uh draft topics to talk about.

C

uh First, is the lsvr flute reduction work, which is you know, being you know, be reintroduced again, and then we have uh two drafts from randy which are speaking of you know uh like l3dl, but then over uh layer, 3 technology itself, the first two topics so for the next step. So where are we with the working group itself and with our work? So so? Finally, you know we were able to request the iana code points for our main specification draft. It has been reviewed and it actually has been re-edited also.

C

uh We are now at dash of version 16, and I believe this is ready to go towards iesg. If I'm not, you know incorrect, is that correct. uh Victor.

C

Okay, it's probably not connected on the mic.

B

Yeah, that's correct.

C

Okay, yeah yeah, so so we'll be doing that you know very shortly so it took you know a lot longer than as anticipated for getting to the core points, mainly because you know the chairs, you know, had to figure out how to request those things, so that took more time than anticipated. So, sadly enough, but we are there now and the draft is ready. It has been reviewed a few times, so we believe it is reasonably uh okay. You know it is actually in a good shape and then the other thing. What we were.

C

What we discussed already during the previous ietf meeting is we will have to reshorter the the svr working group and we had to wait upon the moment. We could, you know, submit or finalize our prime draft here the bgp spf and ship it towards iesg, so that we actually have our work. You know our main work item finished from the working group and that could actually kick off then like a reshorter of the lsvr charter itself, and the main goal would be to actually include the l3dl work as formal entities of the working group itself.

C

So this is this: you know this discussion. We will actually kick off. You know starting from next week, and we will start you know discussing like you know what and how the new charter is going to be and then check with the working group. You know if there's like interested people on that one and and to form you know the the correct text for that.

C

Any questions on that or remarks, otherwise we will move towards. You know the draft presentations.

C

Okay, nothing, I hope everybody's microphone is working though, so let me go to the first one. uh So hi mo, are you ready? Do you want me to drive these slides or will you be driving the sites.

G

uh You'll drop the slice. Thank you.

C

Okay, here we are.

G

Hello: everyone I'm going to talk about the bgp spf pilates reduction next page.

G

At first I would like to thank ac video, the donut for their valuable comments and suggestions to the draft. We made some changes to the previous version.

G

Firstly, we simplified the draft through removing a complicated option, which is called the group option from the rr model and then some editorial changes next page.

G

So this is the current simplified, revised floating procedure in root reflector model, so basically, basically bgp spf speaker will send it linkers linked information to one or two neutral reflectors. Even though we have a multiple root root. Reflectors in the network, after receiving the link state, the router reflector will just send the links to the information to the other bgp speakers.

G

So this is simple, for example, in the network here we have two root: reflectors r1 and r2. We have number of nodes a b c and d in the network.

G

Node a will just send its link states to one neutral, reflector r1. It will now send its link states to rr2 after receiving the link state from node, a r1 will send the link states to the other bgp spf speakers. So this is symbol and also the link states. Flooding is reduced next page.

G

The revised flooding procedure in load of connection model is not changed in the draft, so this is procedure is similar to the one in the igb flooding reduction. So every node has a flooding topology in the network and then every node will send the link states information to its peers on the flooding topology.

G

So, for example, we have a real network on the right and, and we have flooding topology on the left. Node a will only send new states information to its peers on the flooding topology.

G

So, for example, node a will only send unique states to node b and then node b, which are on the flying topology node. A will now send the link states to note the c, even though node c is appear of node a on the real topology, but the node c is not here of node a on the flying topology. So in this way we reduce the link status lighting in the network.

G

G

So I think, that's all I think for uh right now the flooding reduction is simplified in the current draft, and also we would like comments and also looks like the draft is simplified and stable, stable, and I would also like adoption course if possible. Thank you.

C

So any any questions towards jaimo or on his work.

C

H

H

I hate to say this, but I haven't read the draft, um I'm assuming the draft talks about how, in the round reflector case, you select one reflector over the other and how the second route reflector eventually gets. The state gets the routes right so because, if there's a failure- and you only send the rods in one place in one direction- then you're screwed right um and I'm assuming. The draft also talks about how you select the floating topology.

G

Yes, yes, that's right! Those details uh already in the draft, for example, for selecting the rotary reflector, so we can select the the root reflector with a minimum id. Those details is in the draft.

G

Yes, thank you. Thank you for your comments.

C

Any other questions from people in the room or remote.

C

Okay, so hi mo, so I think one of the questions you put here on this slide also isn't always about adoption. Now, frankly, I think it is. You know it's a bit early on that. You know before we adopt. You know this kind of work. I would like to see some discussion actually on the email list about this and and to get the sense you know of the people who are going to be contributing to the document you know in the future. You know if they find it like.

C

You know if it is like something that they find useful or necessary for this particular environment.

C

You know just basically to see, like you know, more discussion on the working group list at all, because I think you know in the last three or four or five months. Actually, you know the.

F

C

Has been reasonably light, so I think you know if you would launch a question it probably will trigger like lots of people. You know paying attention and giving you feedback on where to go and how to do it and and if not, at least you know, it's actually a first step into the direction of you know for for having the group working on this particular work item. Okay,.

G

Yeah, okay, thank you.

C

Okay, then, let me jump to randy's slide. You have anything else to say victor for about this.

H

Mission, thank you.

C

Okay, okay randy, you have full control of the deck.

D

Hey hey gunter,.

C

D

I got in the queue late. This is a c lindem, cisco systems. I was just gonna say in terms of the wemo draft. I I've I've read it uh briefly. I haven't gone through all the corner cases, but we always. We always implied that we could do this, but we you know we kept it out of the base specification. So it's in line with our work.

D

I mean that you you, you could reduce the flooding.

D

And set it without a scope in the base specification, so I think it it's definitely in the scope of work.

C

Okay, cool. Thank you.

G

Okay, thank you.

A

Okay, this draft um excuse me this was- has not been customized for lsvr.

A

When idr working group looked at l3dl, they said hey how about also a layer 3 solution, so this is a cousin of l3dl customized for idr discovery, egp discovery in the data center. It's really two drafts the usual suspects.

A

It's attempting to be very boring and uh one bit of philosophy throughout is it's not probabilistic right, you'll see everything is lockstep and you know it immediately switches to hopefully tls, but at least tcp, etc, for just dead, lock, step, reliable protocols.

A

Okay, it's like l3dl, except it's layer, 3., the pdus will look very familiar to you.

A

It uses one udp multicast hello for initial discovery and that's to find candidates for establishing real tcp.

A

It's session oriented and resumable like l3dl, there's no retransmission needed, because tcp does all that it keeps the minimal needed state. So it's not re-transmitting. It's not refreshing.

A

The ends are supposed to remember what they're doing, and the protocol only sends changes due to the fact that it's over tcp tls, it's reliable in order, delivery, etc.

A

Okay, the goal is to find neighbors learn the I layer, three addressing and bootstrap bgp same as before we go through l, you'll notice, little stars once it has discovered the affy saffies, there's a little star in the blue up arrow because it doesn't have 7752 um bgpls communication to bgp.

A

um There's some discussion of that going on in idr. The feeling seems to be that um it's that happens all under the table inside the implementation. I have some questions about that which I'll say for the end.

A

Remember this isn't a routing protocol, it's just to discover layer, 3 addresses on point-to-point and multi-point.

A

Type length value, one-on-one, um we've got version number in here, there's a change. um This is the udp hello.

A

It's about to change, because discussion in idr the flags is to say hey what flavor connection do you want tcp or tls self-signed or ca based certificates? This should really be an enumeration which is raw, tcp, tll tls sell, signed, tls ca based and what really should be? There is tls ca based with address in the certificate, so it's locked to an address.

A

um There's a port in there in the hello. Just in case you wanna, didn't like the iana assign port.

A

The tls session is opened by the hello peer with the lowest ip address.

A

The highest address acts as a client there's some discussion of how this is handled in an internet exchange that has route servers.

A

Two comments about that so far is one that wasn't in the goal set. This is for a data center and two. Is that all the route servers I know- and I do actually run routers that are on exchanges- are at the lowest addresses, so just don't field a server and you'll connect to the route servers.

A

Tls the hello set, whether it's ca based or self-signed,.

A

If it's self-signed, we call this trust on first use. I jokingly call it married on first date, um you've seen it's gone out of fashion. Thank you to lessen crypt and free certificates, but um you certainly remember browser saying I don't know the certificate: do you want to trust it anyway, etc, etc.

A

So it's believed without question by the client. You do get integrity because you've got tls and if you disconnect and reconnect you know, you're reconnecting to the same attacker.

A

If that is satisfied um in general, we expect ca-based keying, which is enrolled in a certificate authority, and the server must use the cert, and the client therefore can check that the cert from the server is indeed signed because it has it's been provisioned with the public.

A

Key of this trust anchor, okay and they're. Actually um we won't go into detail here, but you can find it in rpki um router certificates. There are two modes of enrolling: one is self-generated on the router and the other is generated in the ca, okay, the choice of which is for the operator.

A

So you now have a tls or tcp session, and you wish to have an l3 neighbor discovery session within it.

A

There's a session id which is a nonce that identifies the session. That will be there and it's the same session until something bad happens.

A

There's a serial number in each of the data pdus, it's think of it as a checkpoint that if you decide to restart you can tell your friend, hey. I've still got the same session id. Can we resume from this serial number please and then there's a miscellaneous collection of attributes um in this case we're suggesting to not have predetermined attributes, but leave that entirely up to the act. The operator, okay, all pdus are acknowledged.

A

The acknowledgement is either of e-type 0. There was no error and you're all happy or air actual errors and the errors can be of various severities and there's an error code in their hint that look a lot like smtp errors and browser errors, etc with a code and a further sub code.

A

Okay, so it's fully stateful it's a session. Prepare, it does have restart, it can be resumed.

A

um One of the co-authors is a major bgp freak, so he manages to argue for all these bgp like features okay, so I want to tell you what ipv4 addresses I have and, as I said, there's a serial number and then there's a count of the ipv4 addresses and for each one there's the address and the prefix length, but there's also flags.

A

Am I announcing or withdrawing that v4 address? Is it a primary or secondary?

A

Is it under layer overlay? Is it a loopback?

A

There can be more reserved at the moment.

A

So, for instance, you might have a fee for encapsulation pdu that says here's my primary address and here's a loopback address and if later this system says hey peer and peers peer with the following address and it hands you the loopback address you know to forward through the primary okay, of course, there's ipv6 and mplsv4 and v6.

A

So now we want to communicate bgp stuff.

A

It's meant for arbitrarily or above layer, 3 protocol so far only defined for bgp.

A

There's, a pdu upper layer protocol configuration it's only defined for bgp, and then it has a list of sub tv tlvs. With the count of the elements in that list, the idea is still to provide the absolute minimal set of configuration parameters for bgp open to succeed.

A

You don't want to duplicate, what's going to be in bgp open, because multiple sources of truth are a rep recipe for death.

A

Okay. So what's my asn?

A

What's my ipv4 peering address, of course, there's a v6 version too. That prefix length is an error. It has to be removed because it's just the address. You know the prefix length from the base: pdu, okay, there's authentication data and there's whether you have gtsm ttl check and that's all.

A

You really need, of course, there's one for an ipv6 period, even though the bgp sec says that the same asn2 devices can only have one pairing thanking robert who's reminded me of this hidden fact: um practices they actually do so you can set up a v6 and a v4 appearing with the same pdu.

A

Remember that the base protocol provided marked loopbacks.

A

So are these features or bugs? I think it's a matter of taste. It's state stateful and restartable. It's rigid and has handshakes and pacing solid confirmation.

A

It's got tcp, it's a based over tcp. If you have bgp, you have tcp, you may not have tls, though I think any modern device certainly probably does, but if you're rolling out a data center, I would think you've got it, but you know you don't have to.

A

If you want security, do you roll your own or shut up and use tls in the security world rolling your own security is a obscene statement.

A

Okay, it provides large-scale property mortgage bgp needs. um The thought is for evpn, but the scaling really, if you're not using it, only costs a few bits in the length fields.

A

You don't actually have to fill 3 000 sub pdus. You just have to allow the length field to be big enough to handle it. That's pretty much it except.

A

I have some questions and issues that we discussed in idr and they're still interesting. How are parameters passed to bgp, okay and lsvr? I think it's bgpls, but an idr, you see the implementers say: hey, that's my business! It's under the table. My bgp speaks to my discovery.

A

You don't tell me how to do it, but the question still stands. How has bgp started restarted and stopped?

A

What's the timing, how do we communicate that a friend have said: hey pack, it all in one ulp cpdu when that pdu's received you have all the parameters, you need start the sucker. If you receive another restart it, uh he hasn't said how it stopped.

A

When is discovery finished, and when should it be stopped, should it be stopped right if you're on multipoint, you probably want to keep going with the hellos, because no somebody new can come on?

A

Okay is the restart ability really needed, and that's my deck.

A

A

Do I see snow out the window.

D

Hey gunter, I don't know if you can see the queue.

A

Hi ac just talk.

D

Okay, ac, linda my cisco systems,.

A

D

Sorry, the first, the first comment I had was you were talking about forwarding you envisioned this protocol, actually installing routes to get it to the loopback address. Or do you see that to be the under the un under the purview of the clients of this protocol like, for example, the latter?

D

Okay? So it it it's like it's like lldp discovery where we said we weren't going to install the route okay.

D

uh The second comment I had is that on this last page, a lot of times like, for instance, if you look at other protocols that give hints uh to the it's it's a hint and you would never stop a protocol based on the discovery protocol. Once you start a session, it's up to the it's it's up to that protocol like, for instance, a good example of this. Of course, it's not used.

D

It's used for liveliness detection, not for discovering the parameters is bfd, so if uh well I mean just because you you just yes will be be a bfd. That doesn't mean you, you would stop it right if you, okay,.

A

Yeah yeah yeah, which is why this is a question right. Yeah remember: this was from idr here: okay, yeah right right here, you've got bgpls, so they can. They have a signaling protocol which they can use to argue.

D

Right, I mean the way I see all these discovery protocols the parameters, change, it's not compatible with the current session, there's something that needs to be done. The capabilities need to be renegotiated based on those it's up to be, it would be up to bgp to to stop or restart the session. That's the way I see it anyway,.

A

C

And me also, you know, but then as one of the contributors, so so something what we see. Also in the igp world when doing discoveries is there is an exchange of, like you know, interest for doing like street bgp. So you know there.

F

C

Some extensions in the hello, you know the ways of doing like discovery that the note actually is capable of doing the you know of bfd itself. So from the moment the decision is built up. That's only done. You can actually start up the bgp or the ics or the ospf and then do a fast fail. It actually fails. So is that something you know which you intend to include in this also, you know to advertise a preference that bfd has to be established after your l3dn neighborship actually is established, or do you.

I

See them as they're not compatible.

C

A

I have no objection to that. Anything which is which is not compatible in a data center, in my opinion, is a big red alarm because something broke in the provisioning system right. So if both sides aren't compatible for bfd, then that's an error character. I think wanted to speak to the point I can smell it.

E

Yeah, that's a really good point uh gunter. I think this solution can benefit from auto, discovering and auto negotiating bfd as well alongside of bgp, and as you do that I think um you can go here.

E

Yeah yeah, exactly the fast fast session deactivate.

E

Thank you. You actually raised a point uh where we always made a claim that said that a good generic discovery protocol mechanism is needed, and here it comes that lung side of bgp. You may need something like pfd negotiation to be in place. Also, thank you. That's all I had.

A

Could you drop that in an email, gunter.

C

Yes, no problem I'll send to the list. Thank you.

E

Preferably on the mailing.

C

A

C

Other questions or observations.

A

Thank you very much.

C

Thank you, randy.

C

Let me go back to the first deck here.

D

Yeah, I I just have one: uh are we still going? Are we still gonna uh progress this protocol in this working group? We are aren't we. This is ac lindem again from cisco systems.

C

That's a question to the chairs. I guess.

D

Yes, yeah the l3.

C

L3Dl, so so we were thinking of you know so so when we do the resharpening, then the idea would be to include l3dl and the the layer 2 technology uh as part of the shorter of the working group itself. So this uh this.

F

C

Actually of of l3dl.

F

C

It's relatively new, I'm not really sure. Yet you know if we, if you know, if we act, if we would like to work upon two different sets here. It is something we can debate, but I have the impression that you know alvaro wants to say something about that also because he lined up very quickly.

H

Sorry, paul I'm going to stand in front of you uh a little authentic uh yeah, so that is the plan, and that is what we had agreed with the idr chairs uh way back when all this started.

H

This was of course a few years ago, and it was before um idr put so much emphasis on discovery as well. uh We had, of course, consulted with them, because um you know what we're doing here is discovering the bgp peers right, so we felt that it would apply to other things. So, yes, that is the intent.

H

um We probably need to do one more round of um syncing and consulting with uh idr just to make sure we're all in the same in the same on the same page, uh but you know barring any any issues. Yes, this working group would then do l3dl.

H

um As for uh the layer, three version we'll need to uh have more conversations around that.

C

F

um I I had a just I mean this is my ignorance on the layer, three discovery: uh hello, uh multicast: how? How is that contained within the network? Are there filters or acls or something that keeps those things from flooding all over the place, or do they just flood all over the place.

A

Two things: one: it's point-to-point links two: it's gtsm ttl cam.

F

Oh okay, all right thanks.

A

um Back to alvaro's point um um this, this layer, three discovery is just here informatively, I don't think we're.

A

I mean if you want it. If the chairs and alvaro wanted in this group, that's fine. It was really intended to be informative here.

H

Yeah, you and I are in sacramento thanks.

A

Right and- and there are some interesting things in it- that some of the pdus were a little further developed than the last time you saw l3dl, so they'll, merge and they'll go both their own ways and idr will decide. Would they want layer, two or layer, three or both, and god knows we're just trying to be friendly and accommodating.

D

Ac lyndon cisco systems hey. I was just gonna comment on the scope of the pdus. uh I have read these drafts, but not before this ietf I haven't read the latest iteration of them. So I don't remember, remember everything, but wouldn't I would think you would just use the all routers uh multicast address, which is link scoped in both ipv6 and ipv4, and then it would never go beyond one link. Are you thinking these are going to go beyond one link? These discovery, packets.

D

I mean beyond hop it's early in the morning.

A

I was thinking sue would speak to that, but if not oh she's left um it's allowed, but that's a configuration option. um It's currently not envisioned to do so, certainly not one way or more than one layer. Three, the question is whether layer two care want to speak, go for it.

D

Yeah layer, two, you get that for you get that for free, I mean with with yeah at layer two so.

E

Yeah that that is this is a scale patella again our case. That is how you would probably want to do, um just as you have it to all routers.

E

D

My apologies for not I I haven't read like I said I haven't read it in the last month.

C

Okay, we have still in the queue or.

C

Okay, then I think we, if there's like no further questions, uh then I think we have reached the end of the session here. So so we have used about, like you know, 40 minutes from the two-hour slot. So you got like uh like one hour and 20 minutes back. Thank you all.

A

Efficient protocols, efficient protocols.

C

Thanks bye-bye see you next time see you in philadelphia, hopefully in real time, live.

E

Thanks, hopefully, we'll see you in philadelphia in person.

C

That's the plan.

B

Yeah, hopefully I'll see you in philadelphia too venture see you victor.

E

In philadelphia.

B

B

B