July 29, 2019
<sarang> OK, let's begin
<sarang> Hello all
<sarang> Logs of this meeting will be posted to the GitHub agenda issue afterward
<sarang> Whoops, here is the current agenda: https://github.com/monero-project/meta/issues/377
<sarang> Let's go ahead with ROUNDTABLE
<sarang> I've been working hard on an RCT3 implementation, integrating some nice optimizations and seeing if it's possible to get key images working as expected
<sarang> This also led to a bunch of other library updates that the code relies on
<sarang> So generally just a lot of coding
<sarang> I also gave a fun lecture on the Enigma cipher machine and the math behind it, which is always fun ;)
<suraeNoether> nice, what was that for? a meetup or something?
<sarang> A friend teaches at a university and thought it would be good for an undergrad crypto class
<sarang> Since the Enigma break is all about permutation groups
<sarang> I should be able to finish up an integration of the BP inner product verifier optimization into RCT3 shortly
<sarang> Then, of course, DEF CON approaches!
<suraeNoether> i have a very dirty branch on my computer with my matching simulations and experiments being worked upon
<sarang> How are those going?
<hyc> defcon - anyone got grasshopper repellent?
<suraeNoether> i've proven all the theorems necessary to prove that the algorithm in graphtheory.py does what i claim it does, and i'm currently testing that my simulated ledgers are doing what i expect.
<sarang> Will be very neat to see the results, and to see if/how they can work with the analysis that Isthmus et al. are doing on Monero and Zcash
<suraeNoether> in addition to that, i have a bunch of MAGIC stuff I need to get done, and my discussions with isthmus' digital forensics work are leaning in that direction
<suraeNoether> i'm thinking the content of this project may take up more than two papers :\
<sarang> Also worth noting that the CLSAG paper has been updated on IACR, thanks mainly to kenshamir[m]'s comments and questions
<sarang> kenshamir[m] has also been working on a Rust implementation of CLSAG/MLSAG using the dalek library
<suraeNoether> hyc: i hear grasshoppers are delicious, i wonder if any of those fancy vegas michelin star restaurants are capitalizing on the swarm
<sarang> Anything you'd like to share on that, kenshamir[m] ?
<kenshamir[m]> Hi, the library is finished, not interopable with monero because it uses a different curve though. The numbers are quite impressive and I believe it is due to the way cLSAG was created
<sarang> aw shucks
<sarang> Will be great to see your MLSAG vs CLSAG numbers at different ring sizes
<suraeNoether> i believe there was a question re: ristretto and multi-exp like pippenger right before the meeting that would be relevant here?
<sarang> I brought it up a while back because we handle different linear combination sizes _very_ differently in the codebase
<scoobybejesus> is that rust code in a public repo?
<sarang> We use no fewer than 4 methods
<kenshamir[m]> The numbers above were for different ring sizes, I may have mis-typed key sizes by accident
<kenshamir[m]> 256 and 512 were for the decoy sizes; if I was not clear
<sarang> A big part of the reason CLSAG verification is faster than MLSAG is because of the introduction of a new linear combination evaluation algorithm
<sarang> kenshamir[m]: can you link them again for the logs?
<sarang> (if comfortable having them public)
<kenshamir[m]> <scoobybejesus "is that rust code in a public re"> Not the cLSAG code, once I add documentation and sanitise it I can post it in here
<kenshamir[m]> <sarang "kenshamir: can you link them aga"> Yep sure
<sarang> What does "without Pippenger/Straus" mean? Simple iterative evaluation of linear combinations?
<kenshamir[m]> Yep exactly
<sarang> Impressive numbers
<sarang> Of course, it's a different hash function
<kenshamir[m]> I just did scalarbase mult in a for loop
<kenshamir[m]> haha yeah that too
<sarang> But if the hash function is the same across both of your CLSAG/MLSAG, then the relative numbers are good
<suraeNoether> something that is both faster and smaller and with equivalent security is a no-brainer for implementation (pending audits)
<sarang> Speaking of this, still in contact with potential auditors, who are moving very slowly
<sarang> Nothing to report on that front :/
<suraeNoether> the improved space allows for a logarithmic increase in verification time without actually slowing down the network (at least when it comes to new nodes downloading the network). judging by these numbers, a ring size of 16 or 32 is no longer like pouring molasses or concrete onto the network
<sarang> OK, any other interesting research to report?
<gingeropolous> re: audits, or roll-out in general. Is this the kind of thing that could be rolled-out in parallel with existing? i.e., have an overlap, where the network uses the existing as the primary / default, but can optionally use the new thing, and then once new thuing is vetted, just prune the old?
<hyc> ^ slow auditors - a lot of people are on summer holiday now
<gingeropolous> or, switch to the new and no longer have to relay / verify the old style during the overlap?
<suraeNoether> sarang: am i wrong in saying that gingeropolous is correct that clsag could be implemented in parallel before mlsag is deprecated?
<hyc> we had a 24hr overlap in the last hardfork for this sort of thing
<gingeropolous> this would be kinda different.
<sarang> I suppose it could be overlapping, provided the fee model supports it properly
<sarang> I don't really see why this would be useful
<gingeropolous> i dunno. i'm just opening the conversation regarding the model that all new things need to be audited
<gingeropolous> i mean, obvi auditing is great
<gingeropolous> but its not perfect
<dEBRUYNE> hyc: Yeah in general, summer constitutes low activity
<sarang> It's good for inspiring confidence, as well as the obvious benefits of catching any errors
<kenshamir[m]> Can Monero benefit from using bulletproofs for arithmetic circuits?
<dEBRUYNE> gingeropolous: Didn't we discuss that kind of model for Bulletproofs or RingCT too?
<sarang> Likely not, with our current tx model
<gingeropolous> yeah dEBRUYNE
<sarang> The scaling isn't great
<dEBRUYNE> I think it was shot down for good reasons, but I cannot remember them exactly :-P
<gingeropolous> i think that was post audit though
<sarang> and we have hash functions, which screw things up
<sarang> Zooko had a slide in some presentation where his team estimated the verification time for a circuit with the complexity of Sapling (Sprout? don't recall)
<sarang> and BP verification was O(1 s)
<sarang> Compared to our current verification time which is probably 1/50th of that
<sarang> maybe 1/100th
<suraeNoether> kenshamir: if we designed an arithmetic circuit to describe a ring confidential transaction language, then yes, but that's sort of what RCT3 and omniring and lelantus try to do… sarang, i think zooko was showing "what it would look like for a Sapling transaction language to be proven in the bulletproof setting" not ring confidential transactions, so it's not clear to me whether it'd be slower.
<suraeNoether> in our setting
<suraeNoether> with the exception of our non-AC-compatible hash function
<dEBRUYNE> sarang: That's verification time for fully shielded transactions?
<suraeNoether> dEBRUYNE yeah, iirc
<sarang> Yeah, that timing was just to give an example of what a production-size tx circuit might look like
<dEBRUYNE> Especially given that, as far as I know, you have to run a full node in order to properly perform fully shielded transactions
<dEBRUYNE> Anyway, I digress :-P
<sarang> Anyway, suraeNoether is right in that RCT3 and Omniring try to bring the BP benefits to specific languages used to prove RingCT-type statements
<sarang> which is why we're interested in them
<suraeNoether> yeah, bulletproofs for use in SNARK-style languages is like… uhm… putting a large-diameter turbofan engine into a Mini Cooper. It's not going to do what you think it's going to do.
<sarang> BPs can be much more efficient for languages built for it
<sarang> which is why range proofs are so efficient
<suraeNoether> moving along
<sarang> ok, other research?
<sarang> Or QUESTIONS, from the agenda?
<sarang> OK then! To ACTION ITEMS
<sarang> I'll be finalizing some things for my DEF CON talk, workshop, and panel; and finalizing some RCT3 integration optimizations
<sarang> suraeNoether: ?
<suraeNoether> oh gosh sorry
<suraeNoether> i was computing a number sorry about that. :P
<sarang> it's 7
<suraeNoether> my action items are: work on sims and the experimenter, and work with isthmus to formalize statistical hypotheses for testing all this
<sarang> Having data from Isthmus's group will be extremely valuable for this kind of analysis
<suraeNoether> actually the number is O(153.58*N), which is the number of bits used to describe the number of possible spend histories at ring size 32 with N outputs.
<suraeNoether> which is nutters
<sarang> This is all assuming no external information?
<sarang> Or other graph-based information on chain reactions and provably-spent outputs?
<suraeNoether> merely the total number of self-consistent spend histories
<sarang> Ah ok, so for a hypothetical graph
<sarang> got it
<gingeropolous> O(153.58*N) … great. now your math is talking in math
<suraeNoether> well it means that for, say, 1000 transactions at ring size 32, there are 2^(153,580) possible spend histories.
<suraeNoether> (N has to be a lot bigger than the ring size for the above formula to hold btw)
<sarang> which is… quite reasonable
<sarang> Any other final thoughts or questions before we adjourn?
<sarang> OK then! Thanks to everyone for participating. We are adjourned