--- Log opened Mon Oct 07 00:00:47 2013
17:32 < maaku> so, wizards: how worth it would it be to have a validation index structure that supports transitive/commutative updates?
17:32 < maaku> (U -> A) & (U -> B) -> (U -> AB)
17:35 < gmaxwell> it's certantly worth it if it has ~no cost. I think our conclusion before is that it prevents you from doing level compression, which for index on txid's is fine, because level compression buys you very little there.
17:37 < maaku> It doesn't affect level compression on disk, just the number of hash operations.
17:39 < maaku> A proof right now would be about ~40 SHA-256 blocks; removing level compression of hashes would bump that up to about 290 SHA-256 block - so *a lot* more CPU time
17:40 < maaku> but my suspicion is that even without sha256 cpu instructions the database is still going to be the bottleneck...
17:41 < maaku> of course those hashes would be eating away at cpu/gpu resources used for ecdsa validation
17:55 < maaku> i guess benchmarking is the only real answer here
18:02 < gmaxwell> sha256 is stupidly fast even without cpu help.
18:02 < gmaxwell> 1us per operation or whatever.
--- Log closed Tue Oct 08 00:00:50 2013