From: Miklos S. <mi...@sz...> - 2014-02-03 12:56:25
|
On Mon, Feb 03, 2014 at 04:57:48PM +1300, Teijo Holzer wrote: > Hi, > > thanks for that patch, I've done some tests with the 'writeback_cache' option: > > tmpfs with 16k block size, 2g total file size: > > dd if=/dev/zero of=/mnt/fuse/run/dd1.img bs=16k count=125k > > FUSE (big_writes) : ~50mb/sec > FUSE (big_writes & writeback_cache) : ~80mb/sec > Native : ~1.5gb/sec > > So the writeback_cache gives a performance boost (~2x), but it is > still quite a way from the 20x-30x gain we get with the > short_circuit option. > > It is still CPU-bound inside the FUSE daemon. Do we know what the CPU is doing? Logs, profiles would be useful. Thanks, Miklos |