Evgeny Kotkov <evgeny.kot...@visualsvn.com> writes: > How faster is a commit going to be with parallel PUTs? Would that be > at least twice faster?
If I take a 250MB file containing random data and commit over http to localhost it takes 16s on my machine. The first 8s is the client running at 100% CPU the second 8s is Apache running at 100% CPU. This is on an SSD that will copy/fsync a 250MB file in less than 0.7s. If I have two such files the commit takes twice as long: first the client uses 8s then it waits while the server uses 8s, then the client uses another 8s and finally it waits for another 8s for the server. It is clear that we should be trying to do these 8s blocks of CPU in parallel to achieve better performance. Our single threaded client will always take 16s but the server could overlap its two 8s. If we were to multithread the client it might be possible to commit both files in the time it currently takes to commit one file. -- Philip Martin WANdisco
