--- Log opened Sat Feb 27 00:00:41 2021
05:14 -!- jb55 [~jb55@gateway/tor-sasl/jb55] has quit [Remote host closed the connection]
05:15 -!- jb55 [~jb55@gateway/tor-sasl/jb55] has joined #utreexo
20:05 < adiabat_> I posted more on leaf clustering
20:05 < adiabat_> https://github.com/mit-dci/utreexo/discussions/245#discussioncomment-414118
20:06 < adiabat_> I think it's actually very practical.  Just we don't know how much benefit it gives, but if it's a significant benefit it's totally possible.
20:06 < adiabat_> kind of complicated so should save for later, but... not much worse than the complexity we've got up to now
20:07 < adiabat_> hm why'd I get a _ on my name.  maybe should reboot :)
20:28 < kcalvinalvin> hmmm so keep two trees and have the client go back to the current tree
20:29 < kcalvinalvin> You'd have to rely more on bridgenodes for this right?
20:30 < adiabat_> no, 1 bridgenode can do it all
20:30 < adiabat_> it's more work for the bridgenode though
20:31 < adiabat_> but actually the proofs for the bridgenode to store would end up smaller
20:31 < adiabat_> but a bunch more time for the bridgenode to generate the proofs
20:31 < adiabat_> that seems like a good tradeoff because if you have some setup where it takes the bridgenode 10X as long but the proofs are 1/2 the size
20:32 < adiabat_> then we can make it so bridgenodes sync from each other and don't have to do the 10X work, just one of them does it to start and shares it with the rest
20:32 < adiabat_> basically a bridgenode IBDing from other bridgenodes
20:32 < kcalvinalvin> hmmm ok. I am a bit against more expensive bridgenodes because the proofgen is a big slowdown with ibd now (with the utxo set caching)
20:32 < adiabat_> in fact I think there's no difference in what goes over the network between a csn syncing and a new bridgenode syncing
20:33 < adiabat_> but it doesn't have to be more expensive, they just don't do it, they get it from somewhere else
20:33 < adiabat_> I think that's sortof the same as IBD now. You can't *generate* the blocks, you've got to get them from someone else
20:34 < kcalvinalvin> Ok I get the general picture. Does make implementation a bit more complicated
20:34 < adiabat_> yeah.  that's the thing, it is more moving parts
20:34 < adiabat_> so if the benefit is not that big it's not worth it
20:34 < adiabat_> but... who knows, maybe it's a huge benefit and proof size goes down 90% or something
20:34 < adiabat_> hard to guess without looking into it more
20:35 < kcalvinalvin> Yeah I guess it's worth a try
20:35 < adiabat_> hopefully we can figure out what kind of benefit it gives without doing much implementation work
20:35 < kcalvinalvin> memttldb had a bug so it kept crashing the bridgenode. Still generating the csv for leaf locality
20:36 < kcalvinalvin> I'm thinking up to block 640,000 is enough?
20:36 < adiabat_> for testing yeah sure
20:36 < kcalvinalvin> It's at 580,000 right now
22:49 < kcalvinalvin> With the utxoCache, testnet takes around 1h30m for bridgenode. Very useable I think.
22:51 < kcalvinalvin> oh wait, I'm running off the accumulator before getRootsForwards so it should be even faster than 1h30m
--- Log closed Sun Feb 28 00:00:43 2021