Hello, When test pvfs with iozone, the performance look not so good, is there any way to improve it.
Using iozone with 4G memory, the test command is like below: ./iozone -e -c -s 8192m -r 1m -+k -i 0 -i 1 -t 1 -w -+n >> iozone_result Pvfs use 1 metaserver and 1 dataserver, set the stripe size to 32 M: setfattr -n user.pvfs2.dist_params -v strip_size:32000000 /mnt/pvfs2/dir And modify the pvfs configure with parameter: TroveSyncMeta no TroveSyncData no But the test result is not so good when using PVFS: (1)PVFS test result Test running: Children see throughput for 1 initial writers = 74920.33 KB/sec Min throughput per process = 74920.33 KB/sec Max throughput per process = 74920.33 KB/sec Avg throughput per process = 74920.33 KB/sec Min xfer = 8388608.00 KB Test running: Children see throughput for 1 readers = 85690.92 KB/sec Min throughput per process = 85690.92 KB/sec Max throughput per process = 85690.92 KB/sec Avg throughput per process = 85690.92 KB/sec Min xfer = 8388608.00 KB (2)ext3 test result Test running: Children see throughput for 1 initial writers = 103121.52 KB/sec Min throughput per process = 103121.52 KB/sec Max throughput per process = 103121.52 KB/sec Avg throughput per process = 103121.52 KB/sec Min xfer = 8388608.00 KB Test running: Children see throughput for 1 readers = 150159.00 KB/sec Min throughput per process = 150159.00 KB/sec Max throughput per process = 150159.00 KB/sec Avg throughput per process = 150159.00 KB/sec Min xfer = 8388608.00 KB Question: When test in multiple-node, pvfs performance decrease with node number(e.g. there is only 40M/s I/O throughout when using 30 nodes as data server), Is there any other configure parameter to improve pvfs I/O performance?
_______________________________________________ Pvfs2-users mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
