DASH Community Weekly, 16 Mar 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: DASH Workgroup Community Meeting 20220316

Description

March 16, 2022 Community Call

A

Meeting, please invite me, I can add comments or if you have a final doc, I can take a look and add few additional comments to this.

B

Okay, so let me make a note to invite you.

B

To the next call, if we have one.

B

And uh thanks for that and great contribution by john, is there anyone else who wanted to talk about this particular piece.

C

No, um yes, christina, we are still reviewing it, so we will provide uh it's my informant video. We will provide more feedback next week because it's uh it's a lot of uh new items to review.

B

D

I just this morning create a new github issue. I maybe if it's already documented great but I'd appreciate a pointer, but one of the questions from microsoft asked to ask to microsoft was: is there some it'd be good if there were maximum time up parameters for each of the you know, syn phase established phase and find a closed phase that could be documented as dash requirements.

D

Okay and if there's typical values that might be nice to know for education, but maximum values are of interest.

B

So is this tcp um closing or aging the connections or just like an aging timer or.

D

I mean part of the fin to closing time we talked about was there needs to be a backup? You know if you never see the act for the fin, you need to delete the state eventually, so there's a dash mandates that I think there'd be a configurable timeout for eni.

B

D

Yeah here, knowing what the let's say, maximum possible value dash requires for that configuration value and, and maybe as an example, what they expect. The typical value might be a useful thing to know.

B

Right right, because here the third one from the bottom, we talked about flexibility for connection closed timeout, to be configurable per eni. To x number of seconds. Is that similar to what you're talking about.

D

Yeah, what what is what a maximum value of x would be useful to know.

B

Yeah yeah, so I have that one listed as a work item in the um in the smaller group that talked about behavioral model. So we are tracking that we got it.

B

I'm just going to put c above.

B

And anything from you, john, since you were the contributor um appreciate it.

C

uh No, I don't think so. Okay.

B

That's awesome. Thank you um something else. um I know that for the this right here, the sci uh intel cythrift, ptf and the extension. So I know that um design discussions have begun over at intel and I think the work is going to be starting pretty soon here and if freshman wants to comment. um That would be. That would be great and if not just everyone know that they're starting work over there.

E

Right christina, um as you mentioned, the design is all the discussions have begun um and all the things that need to be worked upon have been identified and the work will be starting uh pretty soon. So we will have a review. We can have a review here um first and then go to the um sonic um tuesday. uh You know community review as well, so yeah we can present here. First uh discuss the design and then present it in the larger community.

B

Okay sounds good, and, and by by work starting, I think what they mean is that the actual coding work will be starting soon.

B

So any any questions or chit chat on that particular topic.

D

Okay, I do have a brief announcement. I could make that would just take maybe 30 seconds.

F

Or so please do.

D

People, um so this is not intel specific. I I host a p4 architecture, working group where we've been discussing the portable nick architecture and um so related. One question: that's come up in the past from about uh nvidia's proposed p4 code is that it uses some non-standard extensions for the uh connection state machine and so, if anybody's interested in those extensions and their behavior and discussion of you know, motivation of why they're created- and you know, brainstorming whether or not they'll be added to people in the future or not.

D

um I'm outing alan lowe here who's volunteered to give a talk next monday uh on this, and I will put in chat the pointer to the how to join that meeting if it's open to the public. If anybody's interested in that topic,.

B

Great great so.

D

That's that may or may not turn into what's in the reference model, but at least if you're interested.

B

D

I just want to let people know.

B

That's a great resource, thank you, and I've been taking your p4 notification and forwarding it to people I think, might be interested. So thanks for letting me know that it's public.

B

So I'll put the pointer in the meeting notes too, and then everyone will have it as well.

B

And so we also um put together a smaller group to work on the behavior model, work, items and p4, and so we have that going. I had a couple of volunteers and if there's anyone else interested just hit me up and I can forward the meeting, invite or you know talk about it on a call or something like that. So let me know I think we're going to meet tomorrow at 9 30 a.m.

B

um Andy, it looks like you went out, did you mean to go on mute.

D

Oh sorry, I I did sorry I'm just um I'm just typing up the details of that meeting, uh so you can put in your notes. Okay,.

B

Sorry, did you ask me a question: I missed it. No, no we're good! um So then yeah. If anyone wants to um more information about that, just let me know, and so I'm going to stop sharing this particular screen and share a different one. Michael and I I'd like your guys's opinion, michael and I were working on a way to track work, and so let me share a different screen.

B

And michael miele- and I we're playing with this idea- it's inside of github and what it is is you know for a certain project, for example p4 behavioral model um specifically marion. You know we had that we had that slide here. That shows the work items that we need to work on.

B

We transformed that into a github trackable um kind of work item list here for this particular project, and so you know we have udp support ipv6 everything that we had talked about. That needs to be done, and then you can assign and github have a status and priority and who's working on it actually, and so this is something we're playing with, and I was curious about your community views around how to with. Is this a good way to track the work, so everyone can see it or would you prefer to work off of powerpoint?

B

I prefer this, but.

A

This is a better way to do. It definitely.

B

A

G

Okay, if I may have uh add christina hello, everyone, the the advantage of these is obviously cindytop, and these items listed items are basically issues and each issue is right. Now are just uh you know not linked to anything, but this issue can be the actual issues, and each issue can is obviously also linked to pr oprs.

G

So everything is linked together and github has the ability to you know, connect the status altogether, so this use will give us a better visibility and better coordination of the work done instead of having disconnected things so and and besides following the status. That is important. Obviously, but it's also the ability to communicate among us and to see where we are, what we are doing and and who is responsible for so we can connect to particular personal persons that are working on the specific item.

G

True work should work. Well, I think, and we we can improve, obviously uh based on your feedback and the way that things are uh will go.

B

Right so it's an idea and what I was thinking, for example, fragment handling. Now that we have a little more information about it. We would go into this project work item and we could link it to the, for example, the issues in the pr's that andy has filed and then they're.

F

B

Kind of together and in one bucket, rather than using labels and trying to do a search with a label. So just an idea um curious what what anyone else thinks.

B

So we can try, it.

B

Okay and then I today we were supposed to, let's see, I'm still working on obtaining the. I know we're working on needing the uh eni sizing and scale uh getting. I think james will come next week, maybe and discuss from sdn engineering uh to talk about that, and then um we still have the northbound interface that I believe it's being translated into sonic yang and from there we'll be able to show it. So that's the work, that's happening there and then michael, and I also made an update to the transformers document.

B

I'll show it to you here there we go.

B

Documentation so with some feedback from different people, we've changed uh this this figure here and this figure. If you scroll down talking about how to determine the direction of the traffic and what we were missing, it looked like what we were missing was we had. You know this information here, but you know with packet direction. It wasn't really explained, you know how we were doing the lookup and then how the eni was selected and and then that double d cap um item that you guys saw inside the p4 code.

B

We we tried to kind of explain it a little more here, but so we basically over on this left hand side if this is the. If this is the sdn appliance- and these are the the dash optimized vms and over here is just external, so I wrote remote endpoints. It could be express router, you know, whatever we tried to make this diagram a little bit more readable, and we put a note here that the vni tends to be static on this side, because the the vxlan here is using the identifier.

B

Usually it stays the same and there is a reserved range that it's pulling from on this side and if I'm going to look those up and see if I can list those here and then over on this side, the the v and I would be different depending on the inbound connection. So we updated this document here.

B

And I don't know if the community sees all these updates or if I'm going over something you've already seen.

D

I see the updates, but the reminder is good.

B

Okay, okay, good good awesome and then um so that's so that's.

B

I hear an echo, am I echoing.

B

Okay, great, so then, um if did we want to, I know that h.a is a little bit further out there, but did we want to talk about um bud's, h8 proposal or is? Should we more be focused on the v-net to v-net scenario and just get that out the door.

B

You see who's all here.

C

What are the open items on the unit to unit.

B

The in the behavioral model.

C

Well, in the behavioral model we know most of them. I thought. Maybe there is something else that is up for discussion. We know what we need to do, I'm thinking about what we still have to agree on.

B

I haven't heard anything new other than the the things we've talked about and and then the request.

F

From questions.

B

Here again from andy, what's up.

H

Sorry that was me with the echo. This is buddy. That's okay,.

B

Oh okay, that's okay! Bud right! So we do have the request from andy to see if we can come up with that time value, but other than that. That's all I know about.

H

So so christina, I have a sort of a general question about the current p4 pipeline code, um like we know that there's still capabilities and features to be added to it. um I was wondering in terms of the existing data structures that are laid out in that code, like how.

H

How confident are we that that code and those data structures are stable like this? What's in the p4 um pipeline code right now, the the tables that are defined, that they are really we're, confident that that is the correct uh layout.

B

um Marian did we base that on any kind of standard design or a typical um people.

C

Yeah and actually right now, yeah and right now, what we're doing we are comparing this to the northbound api to see if there are any gaps in misunderstanding of the behavior, uh so we found some minor differences that are still discussed, but the main flows, like routing all of the mappings validations, seem to be pretty much. What was expected from the northbound.

I

Hey um mary- and this is jay- I had few questions. uh I sent you an email. uh Is there a pr open where I can put my questions on the behavior model and the apis.

C

So apis are in the side, dash branch.

I

In the workflow, which you have described in the pr.

B

I'll show you here, yeah.

F

I'll show you.

B

So if you go to the main section here, there's a serious pipeline and then there's psi and so here.

C

B

C

Be substituted with the tool chain that auto generates those apis and um the there is uh first, uh let's say: first official version of the psi api in the psi. Repository branch.

I

C

I

But the tool chain, I had questions right because you made the claim that, in the last meeting uh the auto generated apis will have the implementation as well uh and there's some yeah.

C

But that's for the simulator only.

I

C

I

That, I think, is not coming out. Clearly, uh I was talking to gohan as well, and there is some some gap there. I just want to make sure that you know at least on the dashboard or somewhere. It's clearly understood. I just want to add comments so that if anybody looks at it, I don't know where the right place is.

C

um Once again, comments to what to tool chain to sci api.

I

C

So so we have a pull request for the tool chain.

B

Let me show you here: go ahead, keep going authenticating.

C

And the best repository yeah pull request number right, pull request: number okay,.

I

I see that okay, so I'll add my comments here and there mainly just just to make sure that we are all on the same page, all right, yeah, all right. Thanks.

B

Does that answer your question bud.

H

uh Yeah yeah, it definitely helps um so along the same topic or similar topic. um Last week I had a couple questions on some of the data structures and I filed an issue. um I don't know if it's uh worth taking up time in this meeting to talk about it, but um is.

B

It the application table and p4 pipeline code.

H

H

So basically, there's a there's. A table in the p4 pipeline code called the um table, appliance, cable, appliance yeah, um and um I wasn't clear like where the index into this table comes from or if it's really a table um in the in a traditional sense of multiple entries.

C

Oh, I see your question, no it it is actually uh the global attributes of the appliance, so it is expected pretty much to have like uh one uh one catch-all entry uh with the global attributes and it it is in reality. If you look at the site api, it is mapped to underlay psi attributes and this mapping will be provided as part of the sonic design, so it based on this table. No new api is generated. It will be translated into the existing site, attributes.

H

C

It's like there is the mac address of the device ip address of the device. It's not really any uh lookup or anything related to sdn processing.

H

Okay and that neighbor max attribute in the table. That's that's a real thing too.

C

um Yeah, that's the taurus, which is mac connected to uh to that appliance.

B

B

All right thanks is it okay, if I just write this down here just to capture it marian.

C

Yeah sure, okay.

H

All right, great thanks.

B

Yeah, is it a okay? Should I just keep this open better? Do you? I can close it later.

B

Do you have a preference but.

H

uh You can just close it down.

B

Oh okay, all right great awesome, thanks, maureen, okay, and uh so we have about you, know about 12 pull requests. um A couple of them are mine and um you know: aha, I wonder if we're quite to the point of doing ha yet with you guys. um This is something we want to talk about today or I know we were talking about how to standardize on the, how we're going to sync the flows between the asics and things like that, or is that for a different, smaller discussion.

B

Well, we can talk about it. Another time, yeah, no, christina.

C

I think a good idea, a good idea would be to again create a smaller working group for each.

G

C

Interested can subscribe.

G

B

Okay, all right, um I can send out a a note or an email asking who's interested, no problem. I'll make a note to do that.

B

Crazy group- okay, great and let's see.

B

So that's about all I had for this week unless you guys have something you want to talk about.

B

I'm going to turn off by sharing here.

C

I have a small update on the site before runtime implementation, so we are now at the stage of uh basically changing the environment. Now it needs to include the p4 runtime library and a different version of the simulator that has this api included. uh So that's what we're doing right now and seems to be going according to the plan by the end of this month,.

B

Cool any questions for marion on that.

G

So can we capture this this work, then, in the in that github list of items correct.

B

Is that correct.

G

If that's, where the okay, perfect.

B

Marion yeah marianne, would it be okay with you if we try to track this in the project format, I was showing earlier uh the update.

C

Yeah, maybe one question about that work. Can I find a link to this tracker.

B

You know this was just something michael and I were doing in a private repo to present to you guys and see if you even liked it. um If you do like it, then we can move it to the main repo I just kind of wanted to get agreement first to try it okay great.

B

So if michael and I move it over to the main repo, then probably by the end of today or early tomorrow, we could have it over there and give you a link.

B

Okay, great well, that's all I had for today guys um unless you want to talk about the the testing framework that we went over last week. um We are going to have to come up, though, with some sort of test to make sure connect. Connection tracking close the connection close is happening in a graceful way that we want it to so at some point, we're going to need to test for that in the in the test, work group, so just fyi.

G

That will allow us to see if the model works right, because this will be more.

B

Right right, so I've engaged porya and I've engaged keysight to see. If that's something we can work on.

B

And so our internal lab, we called the star lab, that's been brought back online, so we'll be able to get it back into there soon and that's about all the updates. I have this week.

A

One quick question: uh I'm I'm just following this from the last two weeks. So are we using p416 rp414 language for this which uh standard we are following.

C

A

D

And if you look at the serious pipeline code, you might notice some things that aren't standard p416 and that's one of the things that will be talked about at a meeting I linked to next monday in the chat.

A

Thanks handy so I'll, yeah I'll take a look at it when, on another question I have is when you're talking about connection tracking.

A

Are you talking about the entire uh thing about connection tracking when you want to implement uh kind of all the exception handling as well as well as the tracking piece of it? When you track, you want to track just the sequence number window size, sync, the cookie data for I flow or tcp window options. So is there any detailed requirement on the connection tracking, just not at high level? Do you have any detailed requirement on connection tracking as well as do you have any exception?

A

Handling like tcp syndrome, re-transmitter window, size, zero or a full retransmit kind of thing. Right are those exception also as part of connection tracking.

D

I I think it's a lighter version than that, like no sequence, number tracking is needed, except perhaps if you want to do an aggressive close at the end, to match up a finish.

A

I just wanted to understand so I'll be working with wenger on this, so I just wanted to understand: is there any detailed uh requirements already mapped out, or is there any request? When I say I see just one line of connection tracking, it could be simple connection tracking or we can go a detailed version of it right. I just want to make sure that we have the requirement. Finally, so that we can plan and execute on that.

B

Marion did we have more requirements other than these that we had talked about.

C

So so this is all described actually in the email thread for the smaller behavioral models thing so uh christina. If you can maybe yeah.

B

F

B

Yeah, so um this one marion.

B

E

You, okay, if I forward.

B

E

Yeah no problem.

B

Okay, all right and do I have your email address.

A

Yeah, it should be yeah. Tomorrow's meeting should have wenger does forward that I'll type in my email address in the chat.

B

Yeah, please do and then I'll forward this along and it's it's a little bit large. But it's because we had to talk about it four or five different times, but um kind of what we were concerned with and where we landed.

A

B

That would be really.

A

B

A

Try to join the tomorrow's meeting I'll I'll, be delayed, I'm pre-booked for that time, so I'll join the second slot. Okay,.

B

All right, maybe all I'll you guys- can ignore it if you want to, but I can also send the meeting invite out to the the larger sonic dash group and if you can join great, if you can't then um we'll just have a smaller group.

A

Yeah, uh hopefully, that will get recorded right. That meeting will get recorded.

B

A

So because I won't be able to join the first stop, so I just want to make sure that if I could I'll try to make it up, but if not at least I want to listen to the recording.

B

Okay sounds good any other questions for today.

B

Nope then I can give time back and some sometimes I'm not exactly sure um some of the members where you're from so, um if you log into the call, maybe if there's a spot to, I know where most everybody's from um I'm just missing. um Who am I missing?.

B

I think it was sandeep sandy. I just get a note that says sandeep joined, but I don't know who that is. Is there sandeep on the call nope not today? Okay I'll, have to figure it out offline. Well, thanks for your time, everyone um I'll give you time back unless there's something else. You want to talk about three two one.

B

Okay, I'll talk to you guys later and please reach out to me um any questions or concerns bye.

G