On Fri, May 06, 2005 at 11:48:52AM -0700, Kris Kennaway wrote: > On Fri, May 06, 2005 at 11:35:29AM -0700, Kris Kennaway wrote: > > > I might be bumping into the bandwidth of md here - when I ran less > > rigorous tests with lower concurrency of extractions I seemed to be > > getting marginally better performance (about an effective concurrency > > of 2.2 for both 3 and 10 simultaneous extractions - so at least it > > doesn't seem to degrade badly). Or this might be reflecting VFS lock > > contention (which there is certainly a lot of, according to mutex > > profiling traces). > > I suspect that I am hitting the md bandwidth: > > # dd if=/dev/zero of=/dev/md0 bs=1024k count=500 > 500+0 records in > 500+0 records out > 524288000 bytes transferred in 9.501760 secs (55177988 bytes/sec) > > which is a lot worse than I expected (even for a 400MHz CPU). > > For some reason I get better performance writing to a filesystem > mounted on this md: > > # dd if=/dev/zero of=foo bs=1024k count=500 > 500+0 records in > 500+0 records out > 524288000 bytes transferred in 7.943042 secs (66005946 bytes/sec) > # rm foo > # dd if=/dev/zero of=foo bs=1024k count=500 > 500+0 records in > 500+0 records out > 524288000 bytes transferred in 7.126929 secs (73564364 bytes/sec) > # rm foo > # dd if=/dev/zero of=foo bs=1024k count=500 > 500+0 records in > 500+0 records out > 524288000 bytes transferred in 7.237668 secs (72438804 bytes/sec) > > If the write bandwidth is only 50-70MB/sec, then it won't be hard to > saturate, so I won't probe the full scalability of mpsafevfs here.
I tried on a quad amd64 machine, which has md bandwidth an order of magnitude greater, but it has the same limiting concurrency of 2.2, so something else is happening here. Kris
pgpZHXGkbkKVb.pgp
Description: PGP signature
