Post-Quantum Interop Call #49

Transcript

Thomas Coratger: Hello, everyone.

Shariq Naiyer: How's it going, Tomas?

Thomas Coratger: Yeah, fine, and you.

Shariq Naiyer: Pretty good, pretty good.

Thomas Coratger: So, let's… Just as a heads up, like, Next week and the week after, we'll not have the call, and then after that, holidays and reorganization stuff. And after that, we will probably reorganize a bit. use the frequency or something, like, I need to figure out, but probably this will change so that we can keep all of the debugging discussion async. And we can use during the calls on various other stuff. But for now, let us keep the format intact, so let us start with client update. Zeam, please?

Parthasarathy Ramanujam: So, I mean, last week, most of the time was spent, running, DevNets, I ran a fork client to DevNet, and we identified that, both justification and finalization stopped, after 7,000, 8,000 slots, so… I reverted to a two-planned DevNet, with 8 subnet of just E, Lambda and z. That progressed well, but as a, I mean, there was some other issues with the servers, so I had to clear out everything and rerun. The problems are still the same, which I've already updated to both, ream as well as, lantern beams. We are getting, delayed aggregated signatures, which is causing this issue. I have received a PR on lean twist from the ET Lambda team, which I have not yet,

reviewed, once, I'll be on it, later today, and then I'll run an Only Eat Lambda DevNet with the latest lean VM, and report back on how, that is functional. That is my opinion. Thank you.

Thomas Coratger: Great, thanks a lot. Then, Reem, please.

Shariq Naiyer: Sweet. We've been looking at the issues that have been happening on the DevNets run by us and Parthas. So, as Parthas reported, our aggregation time was above 2 seconds, while other clients were doing it at near 500 milliseconds, especially Ethlambda. So we're seeing how to optimize it to match that, and I believe that will lead to slightly more stable dev nets. And on the other hand, this may even be the move, but I'm… we're integrating the, the main branch of Lean Multisig, and seeing if that… that just changes things. So, we have… it's working on Ream right now, but we haven't tested Interop. Once we test Interop, we will merge it and we'll notify everyone.

Thomas Coratger: Thanks. Lambda?

Mihir Faujdar: Yep, thanks, Thomas. Hi, everyone. So, yeah, basically, we have fixed the syncing issues, for Lantern that were surfaced, during the multi-client interop run, that was managed by the Zeam team, last weekend. And, we recently implemented, support for the main branch of LeanVM. And we've been testing it locally with IT Lambda. We've noticed improvements, significant improvements in stability. Like, for example, like, the final D slot, it has consistently remained 3 slots behind the head slot. before we implemented those changes, it used to be in double digits. We ended up testing with 6 nodes on a single cell with 2 subnets. We're still cutting, currently waiting for 2 additional servers, which will allow us to scale, more nodes, and we would also like to test with other clients once they have the,

main branch. support for main branch of LeanVM. Thanks.

Thomas Coratger: Great. Thanks, Ethlambda.

Pablo Deymo: Hello, hello? We finished, merging the server spend events APIs. We also merged the touring we showed it last call to visualize how messages arise in a slot. We did multiple runs in, 64 node devnets with egg subnets, all, ethereum nodes. Running this with our latest image was a bit unstable, oscillating time to finality between 20 and 100 slots. With a temporary window of, 2,000 SLOT of non-finality. We did another run with the latest Lambda version, the one that Emil shared. We showed the instability improvement, bringing time, the finality down to less than 20, 25 slots all the times. We stood some… We still have some reorgs on this chain, with 3% of logs being reorged.

In the last 50 hours. We think this might improve by reducing the number of nodes per server. We are running… animation on DevNet, or… 512 validators in a single server. By running multiple Validators from the node. And the specific configuration was 32 nodes in 4 subnets, with 1 Originator per subnet, and 16 millimeters per node. So, that's it from our side.

Thomas Coratger: Great, thank you again.

Bankat | Gean: Thank you, Thomas. Hello, everyone. So, from our side, from last week, we shifted our focus from the operational hardening to prepare Engine for the newer Lin VM. So, over the past days, what we've been doing is, the migration work on our side. We've integrated the new LeanVM signing and aggregation flow. Although that is on an experimental branch on our end here, not committed yet. And then, Obdata declines to match the new formats and, encodings. And also verify that Gean continues to build and pass its full, test suits. Although we've deliberately kept the work on a separate branch, so it doesn't, affect our current participation in the multi-client, Interop, since that is still running on the,

existing, leanVM, variants. So, we also… so that brings us to another question. We don't know if it will be helpful if we could, agree on a single leanVM version. I don't know if that has been put in place already. which lean VM version other clients are using, so that from there, all we need to do for us on Gean is just, pretty straightforward, I think. We generate the required keys, and then refresh, test vectors, and also rerun inter… interrupt testing, so… Basically, it has the updates from James' site.

Thomas Coratger: Great, thank you, Grandine.

Mercy Boma Naps-Nkari: Hello, good afternoon. So, we released, an image last week concerning some optimization and, metrics, fixes. Then, on about updating the, leanVM commit, we're kind of, like, doing that and then testing it out locally. We've been running it for the past, 3 days to see how far it goes, and then the optimization improvement. So, we'll probably, round up that this week, and then we'll make an official image release based on that. Thank you.

Thomas Coratger: Great. Thanks, Now, maybe, like, Emil, you want to give… I don't know if there… yeah, you want to give maybe a small update on, everything related to LeanVM, what you are doing with the binary field stuff.

T. Wambsgans: Hello, everyone. Yeah, so, mostly working on LeanVM, the binary version of LeanVM. Now, we have… a version with a big extension field, which enables us to have small proofs. Still not on the main branch of LeanVM, but that's the… the next… there is still a lot of cleaning and rework to be done, but I hope that we'll be able to merge this big extension field branch to mine, and then, basically we'll have to… continue, like, ideally 3X performance on this. That's it for me, thanks.

Thomas Coratger: Okay, so I can give a brief update about, like, everything related to… speaking and research. So, about spec, we worked with, the Content Respec team to… to transfer the SSZ specifications that we have in LeanSpec, and that was pretty good, into, the official Ethereum organization, so this is still an ongoing process. We have initiated the transfer yesterday, so if you check there, let me share the link within the chat. There, you can see that you have this new SSZ specification on the Ethereum org. So this is exactly the one that we have on, lean spec. So that was done by Justine Tragia and myself, so that this spec will be used among other specifications, including content to spec and potentially exhibition spec.

So that is the work that we have done in the past couple of days on the spec. On the research side of things, we are still continue exploring binary fields techniques, so as Amit just said, like, we have this ongoing LeanVM We speed up and cleanup, like, ongoing. And we have also, like, the… BNUS project that we are working on with Jim, who is an expert in binary fields, of course, to check how we can improve a bit the… the… the proving system and iterate a bit faster with direct… by directly interacting with the circuit instead of having a VM. So, yeah, we are constantly pushing improvement on that. If you have not seen all of that yet, I encourage you to…

to check, the repo, I can share that in the link, in the live. there, so that if you don't have it, you can check. There are a lot of ongoing PRs and stuff in the pipeline, but all of that is in the process of being improved and merged. about, our DevNet, so if I understand correctly, about, like, all of your updates, so most of your client teams, you have tried, like, multi-client DevNet and stuff, sometimes this is a bit buggy. What feels really encouraging is that, we have seen this week, like, updates Especially from, like, the Ethlambda and the Zeam team, reporting, like, more stability when running on a single client, so… what I encourage you to do is that if you have some… some bugs, or some stuff, like, in a multi-client DevNet, so maybe this is something good to just test with your client, and just run, like, a…

a single client DevNets, with just your client, and, like, try to interrupt between your client and your client, so… first you see if this is stable, and then once that is stable, maybe you can start interacting with other clients. I have seen quite good results for that, especially, like, shared by Ethlambda this week, so… Yeah, there is also maybe ongoing tests with the latest version of LeanVM, like, so for… as a reminder, like, before, so before last week, we were testing everything with, let's say the big version of XMSS signatures, so the one that was in a specific branch in LeanVM with a large signature size, and so longer proof time. But Emil just proposed to, so he updated, like, the main LeanVM branch.

Where we have, like, a sub-MTU variant of XMSS, so that is a smaller XMSS. We don't have a proof yet for that, but that is still good to test, because the XMSS variant is much smaller, and so the prover time is much faster as well. So… I encourage you also to test it. I have seen that Ethlambda and Zeam are testing this out. And long-term as well, because I have seen, like, a new image being published for the LeanVM main branch. Be careful when you do that, this means that you will have to change as well the signature scheme. So, instead of… instead of having, like, the lean SIG repo with the implementation of XMSS, you have to switch

to the one that is directly implemented into LeanVM, because the prover changed, but also the CNA should change as well, so be careful about that, otherwise you will have a problem of compatibility, and this will crash, for sure. So… Yeah, I encourage you maybe to do these tests. Maybe you guys have some comments about simulation and stuff, or questions, or things like this. Okay, so… Fine. So, if you don't have any more questions, like, for me, this is good. As mentioned, again, like, I encourage you, like, to… to check, like, with first one client only, or your client, so that you… because I have seen that in your updates, you mentioned, like, problems.

In a multi-client, so I encourage you deeply to check with your clients, and then maybe, like, try other clients if your client only works. So, yeah, that is my advice, and then, as a second advice, maybe test out, like, the main branch of LeanVM, where you will have, like, an additional speedup. So yeah. And, yeah, go ahead.

T. Wambsgans: about this. Lambda reported that, like, when they use the setup-prover, methods that essentially activates the bump arena allocator, which is, an approver side optimization. But it… I just wanted to warn you, there is a drawback. It uses more memory, so Lambda said that with 8GB, it was a bit too short, so you probably need 16 to make it work nicely. But, yes, let me know, like, in case of memory, issues, I guess we can try to… Like, let me know if we need in the API to… a way to activate or deactivate this, Boma allocator.

Mega | Lambda: Yeah, that would be nice. It's mostly for… like, in real scenarios, you probably have, like, 16 gigabytes. We are running multiple nodes in a single server, so memory adds quite fast, and with 16GB, I think we can run, like, 4 nodes. So…

T. Wambsgans: Okay, I mean…

Mega | Lambda: It'd be nice to…

T. Wambsgans: what you would like is a nice way in the API to have, like, a low memory mode and a high memory mode, if I understand correctly.

Mega | Lambda: Yeah, yeah, yeah.

T. Wambsgans: Okay, we'll try to do this.

Mega | Lambda: Thanks.

Thomas Coratger: Okay, so… great, guys, so… enjoy your holidays, and see you around. Please don't hesitate to post debugging stuff in the Telegram channel. Yeah. See you, Ron. Bye-bye.

Parthasarathy Ramanujam: Bye.

Mihir Faujdar: Thanks for following.

Pablo Deymo: Thank you, my wife.

Shariq Naiyer: Boom and day.

T. Wambsgans: Bye, bye.

Transcript adapted from ethereum/pm, CC BY-SA 3.0.