[Ocfs2-users] ocfs2 performance and scaling

[Sabuj Pattanayek]

> Try mounting Ocfs2 in writeback journaling mode:
>
>        mount -t ocfs2 -odata=writeback /dev/XXX /mountpoint

Yup:

> /dev/mapper/vg-ocfs2_0 on /export/ocfs2_0 type ocfs2
> (rw,_netdev,localalloc=16,data=writeback,heartbeat=local)

> By the way, are you only timing the dd, or are you doing dd;sync and timing
> the entire operation? The latter is a better measurement of how long it
> takes to get data written to disk.
>        --Mark

I have a pl script that does start timer, dd, sync, end timer using
gettimeofday() but this seems to give slightly slower results than
doing it like this:

pattans at orca ~/san1/tmp $ time { dd if=/dev/zero of=testFile.porpoise
bs=4k count=500000 ; sync; }
2048000000 bytes (2.0 GB) copied, 15.047 s, 136 MB/s

pattans at orca ~/san1/tmp $ time { dd if=/dev/zero of=testFile.porpoise
bs=8k count=250000 ; sync; }
2048000000 bytes (2.0 GB) copied, 12.5018 s, 164 MB/s

pattans at orca ~/san1/tmp $ time { dd if=/dev/zero of=testFile.porpoise
bs=16k count=125000 ; sync; }
2048000000 bytes (2.0 GB) copied, 13.7218 s, 149 MB/s

pattans at orca ~/san1/tmp $ time { dd if=/dev/zero of=testFile.porpoise
bs=32k count=62500 ; sync; }
2048000000 bytes (2.0 GB) copied, 13.647 s, 150 MB/s

pattans at orca ~/san1/tmp $ time { dd if=/dev/zero of=testFile.porpoise
bs=64k count=31250 ; sync; }
2048000000 bytes (2.0 GB) copied, 11.9441 s, 171 MB/s

pattans at orca ~/san1/tmp $ time { dd if=/dev/zero of=testFile.porpoise
bs=64k count=31250 ; sync; }
2048000000 bytes (2.0 GB) copied, 12.083 s, 169 MB/s

pattans at orca ~/san1/tmp $ time { dd if=/dev/zero of=testFile.porpoise
bs=128k count=15625 ; sync; }
2048000000 bytes (2.0 GB) copied, 11.9659 s, 171 MB/s

pattans at orca ~/san1/tmp $ time { dd if=/dev/zero of=testFile.porpoise
bs=256k count=7812 ; sync; }
2047868928 bytes (2.0 GB) copied, 14.3243 s, 143 MB/s

Anyway to change the default block write size similar to using wsize
and rsize nfs mount options? Still working on the stripe, need to get
some other RAID hardware stabilized.