PQ Ethereum Workshop - Day 3 Recap

Transcript

Okay. Um, we're going to wrap up day three. I can't believe how much we've done over the last few days. Um, I'm told that we're going to have a demo from the de. So, I'd like to invite you uh to uh to come do your demo. So while the student is setting up um a quick announcement again so we can she client internally right now. So thanks to Camil managed to fix all the bugs now

and so is able to talk to all. Yeah, it's incredible and it shows that we put good foundations um you know from the beacon chain of the industry. Uh yeah so my name is Jun from green team and I'm going to show you a quick about two minute demo for current dev and quick disclaimer we in current dev zero we don't have the uh post quantum

signature integrated we are planning to integrate those after dev one but I think dev zero the main focus for dev zero is uh literally interrupting between three clan teams so uh from kin and green and green. So, uh this this is just a simple docker compost script and we have six nodes uh which consists of two uh two for the each client. So, two rim node, two zip nodes, two clean node and yeah I can show you real quick.

Maybe I can clean up. Um yep. So uh I took this generated genesis uh and with this genesis the de will will speed up and yep. All right. All right. So all right six nodes are now uh uh started and we can just quickly look at what's going on right now. So, so there's a six nodes and there's a two another containers that are uh uh under construction of the uh explorer for the lean feature of that. So we are waiting for genesis

- 5us 4. Yep. 16 second left. Yep. And we are using a slightly modified 3S that have been uh written by Vitali few months ago for the consensus. And now we have Genesis uh started. It's hard to see some logs and maybe we can wait a few seconds to get it finalized.

So you might see um Just a second. All right. Yeah. So, we can see just this uh load from Zen team and yeah. So, we are getting five linesified which is very cool and I can show you a real quick about the lean view app which is a explorer. So, it's just a simple dashboard. So, um yeah. So we have six nodes and there's a round robin for block block processing.

So which is just a modular six for each slot and yeah so for link 0 and one is ring ring is producing block as well. Uh 3 4 Z five uh no 2 3 4 Z for five um clean. So yeah it's I think it's going well. Let's go back to the log. Yep. It's it looks great, right? Yeah. Yeah. So, I think it's now uh it's fairly easy to it's this is the cornerstone for integrating those post quantum signature and also snark

aggregation. So, yeah. Uh I hope this can uh accelerate this material. Thanks. So two of the pillars of the lean theorem are postquantum and classic dynamity. And ironically during the postm workshop we have the class finality too button. Um and just to give a little bit of context today it takes about 15 minutes uh to get finality and here is three slots. So 12 seconds massive improvement over what we have right now. And The part of the thinking for the is that getting to 3 SF or even 2 SF is something that will be extremely

difficult to do incrementally. Hence why it makes sense to batch it with the with the post quantum upgrade. Both of which are very invasive uh upgrades that require some some significant. Okay. Um anything else from the death team that you'd like to talk about? any Facebook networking updates? >> Not specifically. I think yesterday um >> uh yesterday perhaps like we've been throughout the week considering different different approaches for broadcast and the different layers. Uh, and yesterday I think we kind of like put all the pieces together as well

chatting with Justin and drawing it all out um on a board and I think it uh this we sculpted out a little initial prototype of what a skeleton uh protocol would look like with eight subnetss, 124 valid agents per subnet, 10% of them being aggregators and uh with a specific protocol for exchanging and enumerating with sign signatures and aggregators aggregates would be available at each point in time. And all of this goes coupled with a rolling or slashstreaming pipeline where aggregates are being constantly produced and emitted uh for then the next layer of aggregators to pick the best match to continue rolling

them into aggregates to then go and and emit a global best match. Um and all of this is would be happening in a streaming fashion which means that both the networking layer and the protocol layer need to be very close together so that we're not leaking performance along the way and this of course brings like percolates into specific design pos decisions as well in the networking stack. Um what you should expect in the next days is a sk the skeleton of a spec that basically outlines the basics of this protocol so that implementers can go and implement it um initially on onp using a very thin profile of lib2p um to then like continue optimizing from

there. >> Nice. Um we also had this this session on on key gen in the morning. Um and basically just to to recap the problem. Um you know we have a lot of secret key material potentially gigabytes of material which is the fully expanded secret key for all of the onetime signatures. Um and there is basically this this optimization which seems to be quite simple where we basically just uh cache the middle layer of the tree which is of size square root. So it's just a couple megabytes. Um and we will recomputee in real time uh the the the

relevant pieces uh for for for a segment of slots and uh yeah the amount of storage that you need is about 2 megabytes and the recomputation time is about 1 second for a whole three days. So every three days you spend one second of CPU time uh to to generate the key. So that's that's a great problem. Okay. Um, next up we have the formal verification. Um, Alex anyhow. I think for today I think it was quite

interesting to see uh the different designs of their CKVMs and the discussion. there as to go see how these are designed actually have can have quite a lot of impact in terms of how easy they are to formally verify. Uh so that'll be interesting to keep following. Um Poseidon to will be quite interesting I think to we're about to start verifying the spfish version of and so we'll be quite excited to then that and see what comes up. Uh I saw there was some bounties and some interest in specifically in in May. So we'll keep an eye out for that. I missed the uh earlier team session we were

tricking with Jack unfortunately unsuccessfully. Uh and uh then proximity them I think was quite interesting as well. Quite interesting as well. I think we'll see what happens with all the various branching out into different projections that typically going onh and end up getting lean one way or another. But I think that's uh presently very future work. But for the week as a whole, I think it was pretty good to one actually be able to gather a lot of people who are working on our favorite in general into one room. Actually, I think it's the first time we've we've had that many people in the same room which has been good and seems as well is presenting coding and then I think a lot

of what gets discussed is always very interesting for us because also been fully verifying it. So having a better understanding of it and having a better understanding of the design constraints uh which then feed back into different challenges oration is also quite important for us and so it's always good to do that and it's always good to have people infor >> yeah um I mean on the duplex front it seems like there's also efforts to uh to standardize this Yes. So KA is trying to push this as an IT standard and so on.

So that's also why we're very to get that one done in market beyond the fact that actually has proof to go down to the which also is quite attractive to transformers. >> So we're trying to push this as an industry standard and do all the work on the foration side of things. >> Yes. So hopefully offices and can formally verify the function much in a much easier way but I think it's quite ready for >> and on the trip basically Alex studed mathematics here at Cambridge. So he was knocking on Timothy Gaus's door and other people's door with the hope that we could find them.

>> Yes. So 13 hours was not here and then uh the other person's information Siri was not here. Uh but I bunk in I bumped into Michael's algebraic topology lecturer who said Tim will potentially be here tomorrow because it's the start of term and so maybe uh Jack won't be here. I'm supposed to be a computer lab but maybe we find some time to go and knock on his door again and uh try to get him interested. I I saw him in June in Berlin so in June in Edinburgh. So hopefully he maybe remembers me. He's going to call me well I don't have a crying kid at this time. Um, >> so one idea that Alex had is for one of

the days uh next year to be in the Newton Institute. >> Oh, we could do the whole week >> or the whole week in the Newton Institute. And I'm told that they have whiteboards even in the toilets. >> Okay, great. Um, next up we have the implements here. So yeah, this morning we had this interesting session about minimal KBM in a nice presentation from Kon and a lot of discussion happening there. So

basically he presented us the paper of M detailing all of the all of the details, how the memory works, uh, everything, the different instruction. We found some very interesting discussion also with Jim from PU discovering that this is kind of similar with the M3 system, the way they handle things and also a lot of synergy with the PR history recussion project, the way they they see the interactions between the various tables. Uh yeah, so this was quite a useful session. This is probably have like an important point in the future of the design that we we could

have in mind for for the lin VM and maybe the future of LINV VM. Um in the afternoon we have also a very nice discussion with two presentation from Angus and and and Rubin from PR3 team and recion. Um Angus did a very good overview of the current status of PL3 highlighting the strengths of PR history that are for example like the current paralization strategy or the current implementation implementation of the small fields like koala bear or baby bear that are currently very well implemented in the finest details like simply optimization and stuff and he

also highlighted some points that could be improved like the for example the transposition or the TFT algorithm. Uh it shows also the numbers that we discussed the other days in comparing comparing like the FFT with the the number of operations that it should require without the FFT. So we found out that there is almost no gaps with small FFT but some sort of gaps with larger FFT. So we plan also in the the coming days and and weeks to check this uh this algorithm and also we spoke about like the future of prompt history involving probably some things uh weird some check

all of this and we had finally a presentation from about the key recussion project why they did this in com because they they had all of the experience with key is a very fast uh recussion things in Chrome 2 and the way they handle like recussion circuit builder. So taking this same approach and modernize it in order to be competitive. So yeah very nice discussion on implementations today.

>> Very nice. Um yesterday I was having a dinner with Angus Ben and one of the comments was that it's all M3 hidden basically the same optimization under the various papers. Uh anyone else would like to talk about the um okay if not we can move to the theory with >> okay so uh yeah we had three sessions today and I think it was a very intense day so started with proid gave an overview of history of proidon and where we are currently with the initiative and crypto analysis I think one takeaway there is that uh we should

maybe pick the constants for the following bounties in in sort of coordination with block 3 and like how they you do the constants uh to make sure that the constants that we pick for the bounties are representative. Um yeah then we had a session on the internet and codings and I think uh this session started already during lunch because I was walking around and seeing people on whiteboard with hashtags and explained the encoding to everyone and then we had a a very uh intense meeting on on these encodings. I think there we had a lot of discussion on how to if we change it like that can we prove security still uh no we can't

how about this and and so on and I think uh there's no definite conclusion but the conclusion is we should uh talk much more about it there are a lot of moving parts and uh one of my concerns is that sort of the coding gets more and more complex and the sort of simplicity of of uh address signatures um is no longer an argument. But I think the the success here was that a lot of people who have not looked at the coding before now know about the problem and got got nerv. Uh and then we had a session on proximity jam which was a very open session and Jako had some had essentially two ideas what to look at

and we split into two groups. One was about sort of the analysis of these conjectures and uh based on the script that Antonio already started. Uh the other group was uh what is still there on the whiteboard is like looking at the conjecture in in very extreme cases like what if you go uh as close to capacity as you can. I think we made a lot of progress there that we now want to write down to make sure we don't use that. And yeah was very great end of this whole week with a lot of theory sessions that were very successful.

>> Yes. So we have at least one paper that was born again but my hope is that we'll have actually severite and I'm sure that after coffee we will have potentially our first result on the simulations which which should be good. Um Jack, you want to share a few words about your perspective on on the progress? >> Sure. I think like very important work in the sense that understanding the regime in which we can hope to prove something and not is very important. So we have some intuition that in this kind of like very problems this seems like to be warranted by the the you know by the proof that we have this attack. So this is very good.

they still you know we we still don't have to do any security advisory because I don't think people you know if you do the extra query for good luck the character doesn't really matter but this is very very good to figure out what can prove and also I'm really excited to see the numeric like some beautiful like new things about parameter estimation so we actually can like start thinking where actually going to prove the conjecture numerically so then hopefully if these two things converge we should be able you get like some much broader understanding be one of the things I would add is like you know these two bars would also talk to each other because when we're trying to prove these kind of things you end up having

like other uh ves the the like statement should be true but then maybe we don't know how to prove them true and luckily we have like numeric we can like try and check it out so I think synergy between the attacks to prove the proofs with gaps and numeric would be very good That's understanding more about the projection. >> Nice. And that will help us I guess define the the rules of the proximity price so that we don't do something that is like immediately this. >> Yeah. And make it harder for us to win a million dollars.

>> Right. We make them shoot all the so for later. >> I see. Um I've also been you know chatting with Neil quite a bit and I know there been like so many optimizations that we found over the last few days. Do you want to share some of the top optimizations in >> no to share? >> But I guess we have a laundry list of like 10 or 12 optimizations and hopefully those compound on each other and will lead to very significant speeds. Um, anyone want to share some?

>> Yes. So actually something that's been extremely useful this week has also been hearing more details about I guess word 2.0 and F 2.0 with speech correlated agreement have been working on uh and so it's been very nice to have these explained to us in person because then we can actually up these into our clip and make our life easier because they're better instructions uh and so thank you to them in particular. Yeah, one thing that I heard is that um thanks to the the mutual product agreement, we have a much simpler proof of pride and that will make the formal verification uh so much simpler. Um so yeah, having Nico in the room maybe say

you know 100 hours bad days or whatever which is which is a great outcome. Um, and you know, I'm trying to think of some of the, uh, ZKVM optimization, but basically like removing some of the twiddles like um, uh, what else do we have? Um, okay. Anyway, I had to hold this out. I'm happy to share it in in in chat. Um, any final words from anyone? Um, one of the disappointments is that the college visit unfortunately um can't happen because the the port is not available, but we will have a visit tomorrow at 2 p.m. So, anyone uh who's still around,

you're more than welcome. Uh you're also welcome to just visit the the village uh you know, in small groups or on your own. And um yeah, to tonight we have the the very special dinner that I've been talking about. Um so, um I'll see you around 6 6:30 for some drinks and then the the talk starts around about 7 p.m. Thank you so much everyone for listening.