DASH Community Weekly, 24 May 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SONiC DASH Workgroup Community Meeting May 24 2023

Description

@oleksandrivantsiv pushed 1 commit.
· c7cc5b5 [doc] Update configuration example for "DASH_ROUTE_TABLE" entries with action "direct" - ask Prince to review

@prsunny pushed 1 commit.
· 72442ee Update dash-sonic-hld.md to elaborate re: a '0' return value shall be treated as no-tag support
https://github.com/sonic-net/DASH/pull/369

PR357 - Ixia controller security risks (TLS 1.1 enabled)
need to disable 1.0 and 1.1, there by default, not enabled purposefully, will be remedied in the next few days

PR368- update saithrift-bld Dockerfile to accept the ubuntu image SHA as ARG

PR369 - update to Scale numbers https://github.com/sonic-net/DASH/pull/369 - Kristina merged - Guohan reverted for formatting

PR365 - Adding support for metering

PR289 - Still needs volunteer - Dockerfile.saithrift-client has a static tag in the FROM instruction

Q: When the SmartSwitch arrives, will we use static routes in SmartSwitch or other? Does dpu need to perform underlay routing?
How will traffic be directed between switch asics & dpu asics?
A: From the dpu, it can hairpin back to the same port.
Route + ACL action (user does not have to be aware)
Optional DASH HLD ECMP hashing in Underlay Routing section

Memorial Day Week - most ppl will be working (except for Monday)

Likely cancel HA Workgroup for a few sessions

Introduction/Welcome: Amith Gorthi SrinivaPrabhakara Narasimha - new intern at Keysight

A

This week we are on our 70th meeting, so I have a I, have an issue with history: I like to record things so I just think it's interesting to do that. I'll go ahead and turn my camera off. So these are the minutes.

A

From last week last week we had both the community call and the behavioral model work group, so I combined both of the notes into the same email um just for brevity's sake, and so you know, we kind of I think we came to a conclusion around the Apple tags, matching bitmap, prefect list, expansion, blah blah um and then in the behavioral model.

A

Work group of PJ presented a huge PR around metering, and so he um also asked for some reviewers to be assigned, so we did, and so that's sitting out here, I hope you all got this communication and then um here were just the notes from last week and the behavioral model work group I just threw the notes down here and added I tried to add a snip of what was changing and just a couple of snips of the different pieces of the pr here that were changing, but for the whole thing and then the lines here, but for the whole thing um you could click on PR 365 and take a peek at it.

A

B

A

Had a pretty good comment last week about be careful of this compounds.

C

A

And so that's what we did last week so hope everyone got that and then this week what I saw was um there were a couple of changes here.

A

Sasha pushed this one here and can you guys see my screen? Okay,.

D

Yeah, so it's a very simple update to the examples of the v-net is not required when the action type is Direct in the route table so for having women. There is a little bit confusing may lead to to any errors in this WSS. So it's better to remove it.

A

Did everyone else see this one come through if you're, following the the repo.

A

E

This is the metering PR.

A

This is the Sonic.

D

uh I think that we can ask Prince to review and upload this.

A

Okay and sounds good, so I'll do that.

A

And then prints also had an update here. I, don't know if he's on the call, quite yet okay. So what did he do? Oh yeah, he just added this one line here: uh zero return value should be treated as no take support so and I said. Thank you.

A

This is the two I noticed from the week.

E

Nd, please: is there a change in the direct route.

A

What was the question.

E

um Just um if we can mention what is the change that has been made in the hld document, changing the direct route.

A

I'm in in this up in this update here or in general.

E

And in this update here, oh sorry,.

A

The previous one Alexandra or the previous one.

A

Alexander this.

D

Is update of the example section so, according to the definition of the action, direct Divinity is not required and it should not be specified in the configuration okay.

E

A

So here 's what you took out yeah okay, then.

A

So other PR's did you guys have brother.

F

uh The second one update to scale numbers it's based on um Gerald the request so background for 200 gig. We- it was 1 million TCP, one million UDP, um now out of total 32 million background will be 15, plus 15 and 2 million will be left for the CPS measurement. Cps measurements should need 2 million active, usually it's two three hundred couple of hundred thousand uh but leave some room. The other one is Packet size for background traffic to be from 1500 to 500.

F

This will increase the number of packets censored by three times um and the other is the flow capacity.

F

um I think it was 64 million and so on. Keep doubling and at all stages flow capacity that reduced by half the requirement. I think those are the changes suggested for the Hero Test.

E

Yeah 32 million sounds good um when I see the first line, uh TCP flows, 15 million. Does that mean we can use TCP? Also at this moment with the um the Hero Test.

E

Or is it just the scale numbers.

F

Hero tests that is contributed in v-net to v-net, has TCP and UDP, but not to take CRC to This Cloud store manovus, um so yeah Hero Test always had TCP and UDP I. Think your questions in the password refer to the excesses test cases that can be running on the CI, but for the actual Hero Test that uses Hardware that always had TCP and UDP.

E

Got it. This is very good clarification thanks.

A

And even though I know, giraffe doesn't get on here very often just to notify him I'm going to add that and I'll bring it up to him later.

A

I'll, take care of that today when I talk to him thanks Marcia. Thank you.

A

Does anyone else want to cover these in line here.

B

Yes, two items all right: one is it's. It's more of an issue. I'll share my screen. Okay, I'll! Stop! Thank you. Save you! The navigation.

A

B

So this one um zegon reported um a few weeks ago about um the HCC controller, supports some older TLS options which are now considered security risks. So I'm working right now on pull requests that will remedy this. um It was. It was just a matter of disabling um disabling TLS, one zero and one one in some standard, golang HTTP server libraries. So it wasn't that we coded in TLS 101 deliberately they were just there by default and no one ever complained.

B

So we remedied that there's a new release of Snappy and- and uh that goes along with hcfc and traffic engine and controller, so I'm working on a pull request to remember to do this. I just wanted to put it out there that will um eventually uh get in the repo, maybe uh the next few days I'll. Have it ready?

B

That's one thing and uh thank you for finding this something we probably should have put in our own tech-in, which we will do now and then um Vincent asked me to speak up on behalf of this PR here update scientific, build Docker file, and um that's this one here.

B

I asked him to expand on the com on the description a little bit so that people understand the motivation, but what he told me was he'd upgraded. He wants to use Ubuntu 222.04 because it comes with some better um compiler optimizations in the GCC compiler for their back end and um when he, when he compiled his own lib Psy and then tried to link it to the standard SCI Thrift server, which is built from the psy repo and in the current in the current Dash make files and Docker files were used in Ubuntu 20.04.

B

You can't just link code compiled in 2004 with one version of GC C live with 2204 you'll get comp, you'll get Linker mismatches, so long story short. He wants to use a different Ubuntu for his builds and he's created. Some changes that allow you to pass in different versions of the Ubuntu um Docker image and that's what this is about.

B

I've asked him to expand the description a little bit, but he's got some make file changes and in the docker file instead of having a hard-coded, Ubuntu 2004 tag, which is here, he's he's allowed to pass in an argument so that you can select what you've unto for your own build environment, and this this actually makes me think that at some point, we'll probably need to improve some of this build environment in general to handle Debian builds as they're done in Sonic management, so this might be kind of a preview of some other changes.

B

We'll need to make Downstream I just wanted to alert people to this whole request and if they would like to weigh in on it and review.

B

Benson also stumbled on another shortcoming which has been in for maybe a year. The build environment was modified at one time to generate the shaw 256 of the docker piles to detect. If there's a change in to rebuild is necessary, there is one shortcoming in that technique and that the docker file might be pulling in some content. That itself changed like some other package and that that Shaw detection technique fails to catch changes in its dependencies, and several developers have stumbled over this problem.

B

The way we've handled it is by, for example, changing some static comment in this file in the docker file, so that the shovel change it's a little bit of an artificial, clumsy approach and um Vincent is is has put in some changes to try to handle that too. So, if people want to take a look at this or weigh in probably take a little bit of thinking to find the perfect way to do it, but I think he's I think it's pretty close um kind of uh low-level annoying details about building Docker files.

B

Etc I just wanted to point that out. If people want to look at it, they're invited to help review.

B

um Awesome, that's awesome and uh I know Andre you've stumbled into some of these Docker dependencies as well.

G

H

Thank you for pointing this out. Yeah.

B

B

um Okay and also um I'm working on.

B

Trying to find the comment here: I I had I'm working on while I'm working on this ACC change. I also came up with a better way of detecting the dependencies um for having the right, Docker images on and actually look for them in the docker Docker images output to see. If the image is there and build it. One time on demand that'll that may come up in this PR, because I gave some feedback to Vincent. I'll also put it in the xcfc one which I'll have you ask for people to review like Andre?

B

So that's it not exciting stuff. This is uh sausage making details. It's Christmas yeah.

A

Helpful, so those were your last two.

B

A

Cool uh prints, I threw something in the chat window for you um since you joined and did anyone else have a PR they wanted to cover.

F

I have a question, not the pr ank. Is this a good time or yeah Smart Switch is gonna arrive?

F

How will there be static routes in the underlay of the dpu AC Kansas Smart Switch, that will direct the traffic to uh different parts of the AC or how will that happen or the traffic is going to be reflected always back on the same port that it came? How will the switch basic communicate with the dpu, Asic or A6, and how will be the internal routing done? Some hardcode is static. Router just reflect back. Anyone knows.

A

Prince I have my guess, but um Prince. What do you think.

C

A

H

Ahead, sorry, what was the question? Oh okay, okay, I can read that. Okay, oh so that, uh basically we can have uh um a different at that slide, so one thing could be like a weekend match on um um on an axle um and then um basically so of course we have the uh the destination Whip or the destination IP address right.

H

So with that, we can determine that it is designed to a dpu and to identify which eni and those things we can match with an axle um for the source or the destination Mac in the packet.

H

So does that make.

A

Sense or I think he was asking.

F

A

Or bgp, or are you asking about the routing protocol.

F

About the dpu A6, so the dpu acid will have multiple, let's say physical ports or connections, how we still know and which connection to send it to the switch async. Will it have some hardcode static routes or it will just reflect the packet back on the same port, internal.

H

Reports from from the dpu it can just uh hair, pin back the traffic to the on the same port. That's at least the thought right now, but from the switch to the dpu like I mentioned, it can be based on a route plus actual action.

F

Okay- and this is this user configurable or this is just internal and the user- has no visibility.

H

Okay, that's right so user doesn't have to be aware of this. When the when the eni configuration happens on a on a switch at the um at the high level, um the switch will determine which TPU to program and then program the switch Asic to have such rules to forward that uh eni traffic to that dpu yeah.

F

H

But this again.

F

Like is there any, is there any HR deal or anything.

H

F

Describes the internal workings of the small switch.

H

That's all so these are all uh currently in uh proposals, so I think we will be publishing a document soon, like wow.

A

H

A

Been I've been um working on it.

G

Just to elaborate on your answer right so when you say we can do here, painting right does that mean that dpu does not need to do any underlay routing essentially I mean just to confirm that.

H

Right so dpu doesn't have to do an underlay routing if it uses happening. So if you look at the underlay routing mechanism that we capture in the Sonic, actually these are the two options. One um dpu can look at the under layout, which is pretty much one single route of default. uh Gateway or you can it can utilize the the hairpinning model, because we we don't expect dpu to do any kind of pcmp hashing. Okay, it's pretty down in the document yeah.

G

It was there, I saw it it's just that there was still some confusion internally within our team, so I think it's good that you're clarifying it, and it's in the note. So thank you. Yeah.

A

Sorry, I'm flipping between a million windows, so okay.

H

I think it's uh yeah this one.

A

A

A

G

A

Yeah, okay, uh anything else: okay,.

H

Can we mention, like optionally, same bashing like instead of noise,.

A

If I could spell yes.

G

If you do ecmp hashing right, that means will you program uh route, so is it like a default route still is it? Is it passed based.

H

On it will be just the default route, yeah.

G

Okay, so these options, ecmp hashing based on default route, is option. Okay,.

A

Okay, cool um any other questions.

A

A

Okay, so I notice, we have somebody new on the call amith. um Maybe we could um introduce you and ask: where are you from I haven't seen you before.

C

Oh yeah, so hi yeah hi, so myself, Ahmed uh I'm, a new intern joint recently in May 15th on May 15th.

A

C

A

Or from a different company, no.

C

No I'm a student actually so, okay.

A

Okay, wonderful welcome yeah yeah, please feel free to jump in it's a friendly group. If um you have questions or anything.

C

A

Yeah, okay, okay, nice to meet you and um so guys, I wanted to talk about. um If we're done with the technical Parts uh Memorial Day I was curious. If you all plan to be in next week for Memorial Day, we usually have a couple of Dash meetings per week or, if maybe y'all, plan to be out or should I send a survey.

A

Okay, I was thinking about being out. What's up.

D

A

Be working too okay! So now, if you guys.

D

Are working then.

A

I'm, if you guys, are working then I'm working so okay, okay, um so most people.

B

You can always take a vacation and hand the Baton to someone else. Everyone deserves a break.

A

Yep I guess I could I guess I could I could coordinate that so, um okay and that's really all the content I had for effort. You guys have something else.

A

Okay, then I can actually give time back if you like so I'm gonna give the countdown seven seconds Andy.

B

Yesterday, I was in a meeting where I said seven seconds. I was counting my head and second six, somebody that said something it works like you like. You just did now.

B

A good one, Andy.

A

I'm going to go ahead and stop the recording okay.