Andrew Toth
Andrew Toth
I benchmarked the latest branch with default dbcache up to 912683. Results are a speedup of 14% - 5:07 vs 5:49. | Command | Mean [s] | Min [s] |...
Did the same benchmark with 5000 dbcache and there is a 6% speedup :rocket: - 4:27 vs 4:44. Even with far fewer cache misses this change is still a benefit,...
Updated the input fetcher significantly: - uses counting_semaphores to synchronize threads instead of mutex + condvar. - stores tx + vin indexes in global vector instead of copying the COutPoints....
Removed `m_batch_size`. Each thread now increments the atomic counter by 1.
> I noticed that for small dbcache it has a very noticeable effect, but for very high dbcache this seems to add an extra cost - since we already have...
Rebased due to conflicts. Removed the first commit. We don't need to modify `EmplaceCoinInternalDANGER`, we can just insert the coins as dirty into the cache. They will be set dirty...