DASH Behavioral Model WG, 13 Jul 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: DASH Behavioral Model WG weekly July 13 2023

Description

2 Intel Engineers (fruffy - Fabian) contributing to TestGen in Issue399
Length of time to generate packets (consider this), run the test to see how long it will take
PINS team is using TestGen at different cloud company
Ping MarianP offline for updates

Andy likely waiting for DPDK developers, compiler issues are open, then Andy can resume BE experiment

A

Behavioral model, meaning today I'm sharing my screen and the notes from last time, I'll turn off the camera. So um the last time we were talking about.

A

We had all these notes from last time, I'm going to switch over to the the issues that I saw come in. um I saw this one come in from frufi and do you guys happen to know who Fabian roofie is.

B

He's from Intel but I, don't know uh what change he made. It.

A

Oh, that's! Okay! No! No! That's.

C

A lot of work in the p4.org and he created a P4 test gen tool and tried it out on the dash behavioral model and came back with some um comments. Good observations about some dead code, so um hopefully the people responsible for that code and I think Marion might be, can look into this.

C

So um hopefully someday we'll make this part of the CI build. But it's a little early to try to do that, but he did it. Experimentally, okay,.

B

Yeah, so this is a test P4 test. Gen is the name of the tool that's been created uh from from Intel fabiography uh is the main developer there yeah before question yeah yeah it tests all the uh Parts in the you know, Hardware or behavior model um based on the P4 program and um uh yeah. It's a test, testing tool.

A

A

um Nate Foster, yes, do you guys know who Nate Foster is because.

B

Whose part-time works at Intel and otherwise is uh yes and Nate, is uh um a professor full-time at Cornell and he's the one who led the P4 test, gen project um and yeah. We are also working with other CSV like Google on this one.

B

It's supposed to be applicable for both switch as well as Nick, which is in our case the ipodpus um yeah and any uh Hardware behavioral model can be can can use. This particular prefer test Gen 2, and it will find that defects. You know if there is uh it, it basically exercises all the code paths.

B

Hardware body parts.

C

um Also, just on Nate Foster's behalf he's a full-time Professor Cornell he's actually co-chair of the people, languages, design, working group and he's one of the leading figures in the P4 community.

A

That was fantastic that they contributed, so I I did see that come through, but I just wasn't sure. I hadn't been introduced to these two. So that's awesome.

A

A riff I know you're a little bit newer. Did you uh have any questions about this? One.

D

Oh I know like there. um There are certain like um tools, then um you know uh later come um Pakistan, because people passages sometimes it's like when the program gets too large. uh It actually becomes a little bit slow for like really large program, so I'm not sure like how long it will actually take to generate all the packet for RPM.

D

uh So the uh and the other question I'm thinking about uh is that there are certain uh programs that comes later, but it's the same idea just like improve the people, so I'm, not sure if we have considered uh any of those, because you know uh it's just trying to improve the performance um but yeah.

A

That's great thanks for thanks for that ref. So um you know in the in the dash repo under pipeline. We have, you know the bmv2 and the PSI, and this is basically what we have so far is certainly not fully fleshed out, um but this is where we are so far so just FYI, yeah and then um oh go ahead.

D

Oh yeah yeah just curious yeah, nothing particular yeah.

A

Okay, all right.

B

uh I, don't know sorry I missed that comment. uh Were you asking about the uh before question? If there is any question, please let me know and I can get some answers for you. Is it related to scale Etc? If you have any questions, please send that email.

D

Sounds good and, uh of course, if you have run the tests like see um how long it will take I would be like uh very interested to know yeah, because.

C

D

Just like yeah academic papers.

B

D

Not yep so just wondering like how the Real Performance looks like.

B

Thank you got it.

D

C

Sure, yeah I, don't know if you read the complete um set of comments on that issue, but uh Fabian was expressing concern that you know for a complicated program. We can try to spend quite a bit of time on um finding all the code paths and you know sincerely deterministic, it's a heuristic algorithm, and so at this point it's kind of premature to say it's something we could use like in I'll call.

C

It quote: unquote: production for dash behavior model, but it's worth keeping up with the progress and Fabian he's working on another program called Flay, which is in his own Branch right here, yeah, which he linked to and that's an experimental project that'll specifically designed to find dead code. So that might be something that we could run. Let's say just as a almost like a linting that, if it's more performant and and not so open-ended, it would be a great tool.

C

Just like running a link check almost um trying to do heuristic generation of test packets on arbitrary programs is quite an art form. We're probably not going to be there for dash for some time.

C

So we'll probably do more explicit handwritten tests, but there are a lot of tools out there in the wild for generating automatic test vectors, I'm, just not sure of any which are suitable for dash right now or who has the time to try to play with them. All contributions are welcome.

D

Yep definitely definitely yeah yeah Google.

C

Google has done a lot of work in that area too. Yeah yeah.

D

They do like a lot of things in the pins to actually do similar. Things to um you know, generate the test package, so they can make sure like seasonal.

B

um Right so yeah Ben's team is actually using before test gen at the moment at Google.

D

Oh nice, nice! That's really good to know that that means and it actually works.

B

Right right, right, yeah, yeah,.

D

I should say Well, it actually works. Well,.

A

Nice, that's awesome, that's awesome! Yep, yep and I know we needed a couple of updates. Today on. There were two that um Marion was working on, however, he's not on the call, so I'll have to Ping him offline for updates.

A

So I guess I can open it up to anyone else. That's what I had seen come through other wait. Let me look at my PR's real quick.

A

Yeah yep, that's that's what I had for PRS and issues. um Did you guys have anything for today.

A

I don't know if Andy's on do you know if Andy um worked on his back end experiment um about the dpda back-end test, experiment or I can ping.

C

Them offline, I think he's waiting for you know developers, dpdc developers to address those issues and they were not high priority.

B

um Some compiler issues are open and those compiler issues need to be fixed and then he'll be able to resume uh the dptc back-end Tax test.

C

Knowing Andy the the microsecond, those fixes available he'll be broadcasting all his progress in my new detail, all.

A

Right, oh, that's, great, okay, awesome, okay and then um thought I haven't seen you for a while. Did you have anything you needed to have addressed um before we close off for today.

A

Okay, given Andy's seven second rule there so um VJ are you good? Do you need anything.

B

uh No I think I'm good this week. Okay,.

A

All right thanks so so that's all I really had and without um Marion's um updates, like I, said, I'll try and get those in like an offline format. So thank you guys for coming, though, and thanks Chris for the context on this test gen. This is really interesting.

C

Yeah and thanks reshma II for the Intel view of it they're doing all the work.

A

Yeah yeah: do you think reshma should I um invite them to the repo or should I just let them contribute, as they have been.

B

uh Invite them to the repo to the to the calls or.

A

No! No! No! No! No, not.

B

A

The I mean you, can you can forward it if you want or I.

C

A

um We can talk about that. um Okay, no I meant to um to be able to write to the repo, but okay.

B

Okay, yes, all right, maybe we could do a phone okay again, yeah.

A

Okay, all right all right! Well, thanks! You guys have a good day. Bye-Bye I'll stop the recording. Okay! Thank you.

C

A

Pre-Friday, okay, bye-bye.

D

Have you pretty Friday? Thank you. Bye.