February 19, 2020
<sarang> OK, let's get started with the meeting
<sgp_> hello :)
<needmonero90> I caught the meeting!
<needmonero90> I would like to note that the meetings are not listed in the calendar
<needmonero90> idk if thats intentional
<sarang> Which calendar?
<sarang> And how are meetings applied to it?
<sarang> Meeting times/agendas are always listed as meta repo github issues
<sarang> Anyway, does anyone wish to begin the ROUNDTABLE with research topics of interest?
<sarang> Take it away UkoeHB_
<UkoeHB_> I finished designing an escrowed marketplace 'protocol' which hopefully solves issues encountered by rbrunner in his openbazaar integration analysis. Also, multisig and txtangle have been finalized.
<UkoeHB_> Finally, I had an idea for reducing minimum fee variability, and likewise for putting antispam directly in the protocol instead of relying on minimum fee
<sarang> Are you seeking analysis on those?
<UkoeHB_> Which is issue #70
<UkoeHB_> They are open for comments any time anywhere
<UkoeHB_> Ah and sarang provided a draft for a tx knowledge proof chapter
<UkoeHB_> (not really my research :p)
<sarang> Heh, it's more of a summary of what's in the codebase (and some changes)
<sarang> I look forward to reading the update draft you linked
<UkoeHB_> A number of topics here are lonely and want attention btw https://github.com/monero-project/research-lab/issues
<sarang> Thanks UkoeHB_
<sarang> Any questions or comments on those topics from anyone?
<sarang> Please go ahead!
<ArticMine> I have taken a look at issue 70
<ArticMine> It actually has serious implications
<ArticMine> When the LT medium increases substantially
<ArticMine> I do have an idea for a solution
<ArticMine> Very preliminary at this stage
<ArticMine> As for an interim fix
<ArticMine> The est is to pay the high or at least normal fee for escrows that are expected to last past the next hard fork
<ArticMine> I will have comments on the issue in the next two weeks
<sarang> Thanks ArticMine
<sarang> Any other questions/comments from the topics presented by UkoeHB_?
<sarang> I'll share a few things
<sarang> First, the Stanford Blockchain Conference is happening right now (and the next couple of days), and has streaming available: https://cbr.stanford.edu/sbc20/
<sarang> Second, I did some math/code related to multiparty stuff for next-gen protocols
<sarang> Third, I worked on code and write-ups for transaction proofs, both for an updated PR and for inclusion in Zero to Monero for better documentation
<sarang> Fourth, I used chain data from n3ptune and friends to do better estimates of the cumulative effects of next-gen protocols
<sarang> both in chain growth and verification time
<sarang> Major caveat: these assume the same input/output distribution as the current chain, and are _estimates_only_
<sarang> and apply to post-bulletproof chain data only
<sarang> ^ this link shows the total chain growth estimates for various protocols with varying ring size
<sarang> namely, from 16 to 1024 in powers of 2 (lines for visual aid only)
<UkoeHB_> Sarang would you mind adding an indicator for MLSAG and CLSAG at the 11 ring size 'point'? For reference
<sarang> Sure, let me grab that data from my spreadsheet
<sarang> hold please
<UkoeHB_> Or the super steep slope from 11 to 20 lol that goes off that chart
<sarang> Heh, I had that data but didn't include it since it's crazy linear
<sarang> I'm running the N=11 code for MLSAG/CLSAG, which I don't have handy
<sarang> Anyway, I'll pull up the time data while we wait
<sarang> ^ verification time estimate for _group_operations_only_ at varying ring sizes
<UkoeHB_> I think it's interesting that all these protocols/signature schemes are similar size on the small end
<sarang> All the verification times are linear (up to a logarithmic term due to multiexp)
<UkoeHB_> Where is tryptich multi hiding?
<sarang> It's underneath Triptych-single
<sarang> They're essentially indistinguishable
<selsta> Does Triptych single have advantages over multi?
<sarang> RCT3-multi suffers due to input padding requirements that still have a linear verification effect
<sarang> selsta: a complete soundness proof :)
<sarang> Update on MLSAG/CLSAG size estimates…
<UkoeHB_> Could you make a smaller graph from 0 to 128 ring size? Since those large ones seem pretty unreasonable
<sarang> At N=11, MLSAG for that chain range is 7.84 GB, while CLSAG is 5.84 GB
<sarang> (the actual size of that chain range is 7.9 GB)
<sarang> ^ same time data, zoomed in
<UkoeHB_> Perfect thanks :) are time estimates for CLSAG/MLSAG available?
<sarang> Heh, just writing that out
<sarang> I have very early estimates on that, which are tricky since multiexp doesn't apply, and hashing is nontrivial
<sarang> MLSAG N=11 estimate is 29.9 hours for that chain range (but I have _not_ double-checked it)
<ArticMine> What hardware was used for the verification time calculations?
<sarang> It's a single core on a 2.1 GHz Opteron machine, with a bonkers amount of RAM
<sarang> I would rely on the timing data only for comparisons, not absolute values
<ArticMine> age of CPU?
<sarang> I am still in the process of getting CLSAG data, which requires additional test code
<sarang> It's a gen-3 Opteron, if that's what you mean
<UkoeHB_> Is there a way others could run the same tests?
<sarang> Again, only estimates using performance test code
<sarang> For next-gen protocols, it's quite easy
<ArticMine> Yes great it does give an idea thanks
<sarang> Well, somewhat easy
<sarang> You need to get multiexp performance timing data and use a linear interpolation that you plug into the simulator
<sarang> For MSLAG/CLSAG you need to run more operation performance data
<sarang> This is the simulator, which is still WIP: https://github.com/SarangNoether/skunkworks/blob/sublinear/estimate.py
<sarang> But again, it's tricky to do comparisons between MLSAG/CLSAG and the next-gens
<Isthmus> (drive by data)
<sarang> Wow, that's quite low
<Isthmus> Oh yeah, the numbers are one thing. But moreso, we should all be more alarmed that analyzing something like this is possible for an outside observer
<sarang> Yep, and has certainly been a topic of interest!
<Isthmus> It's a privacy risk to use subaddresses right now…
<Isthmus> Anyways, I gotta bounce, sorry to spam n run
<sarang> OK thanks for sharing the data Isthmus
<sarang> Another good reminder that I/O structure reveals some information about subaddress use
<sarang> Since Isthmus had to leave, were there other questions/comments on the data that I shared above?
<sarang> UkoeHB_: if you want to run tests as well, let me know after the meeting and I can let you know how to get the numbers you'll need
<UkoeHB_> My computer is quite weak, was just asking for viewers :)
<sgp_> sarang: can you remind us on the plans to fix this subaddress thing?
<sarang> ah ok
<sarang> Requiring separate tx keys per output is a good idea, but IIRC didn't have a huge amount of support when last brought up
<sarang> FWIW the size data that I presented for next-gens assumes a separate tx key per output
<UkoeHB_> Is that necessary for the protocols?
<sarang> For the proving systems, you mean? No, not at all
<sarang> They don't care how you get signing keys
<UkoeHB_> Can you estimate the amount of additional pub key data? Num outs * 32 and num tx * 32?
<sgp_> sarang: why did it not get support now? complexity? size? verification time?
<sarang> My numbers for MLSAG/CLSAG include separate tx keys too!
<sarang> Also: n3ptune's dataset includes the pubkey counts
<sarang> So I could run that separately for a more direct count
<UkoeHB_> With only 3% subaddress adoption, the difference is likely on the order of 100MB
<UkoeHB_> Or 2% of total size I think
<sarang> that's probably a good order-of-magnitude estimate
<sarang> But IIRC scanning requires checking all pubkeys
<sarang> So either there needs to be a specified correlation, or there's added complexity in scanning
<UkoeHB_> I think it costs ~1GB for 30mill pub keys btw
<sarang> I think moneromooo had a better idea of the impacts, when it was brought up earlier
<sarang> FWIW I think it's a good idea unless it's very compelling not to due to complexity
<sarang> OK, we're running up to the one-hour mark…
<sgp_> obviously without this change, the impacts are quite negative for network privacy……..
<sarang> It's differentiated data, but it doesn't leak _which_ outputs are subaddress-destined
<sarang> (not that I'm saying that's a good reason to keep the current approach)
<UkoeHB_> It's quite a lot of unused data, I'm a bit skeptical
<sgp_> just reveals "one of this outputs goes to a subaddres?"
<UkoeHB_> A lot of dummy data
<sarang> sgp_: it reveals the number of subaddress outputs
<UkoeHB_> sarang all it reveals is at least one of the outputs must be to a subaddress
<sarang> Doesn't it reveal the total number of sub outs?
<UkoeHB_> How would it?
<UkoeHB_> Number of additional pub keys always equals number of outs
<UkoeHB_> Even if nonsubaddress
<UkoeHB_> How is the CLSAG paper going?
<sarang> Hmm, for some reason I thought otherwise; noted
<sarang> I'm still waiting for suraeNoether
<sarang> He wanted to continue working on his ideas for the security model
<sarang> So unfortunately I am not the one to ask
<sarang> OK, is there anything else of interest to share?
<sarang> (Would be a good idea to continue discussing this after meeting, or on an issue, to keep it alive)
<sgp_> definitely need an issue for it
<sarang> All righty then; thanks to everyone for attending today
<sarang> We are adjourned!