Mark Bentley wrote:
I'm already spreading disks in a volume across the 7 SCSI controllers, but I'm wondering if there is also a performance advantage/disadvantage to spreading out the disks in a volume across the three IDE-to-SCSI bridges in each R100 disk shelf?
yes, per my testing. The scsi-to-ATA bridegs appear to serialize transfers. Also, ATA bridges can fail, so IMHO you want to arrange your raid-groups so that any ATA brige failure can only take out one spindle in a raid-group, and the minimum number of spindles in any given volume.
What I ended up doing is somthing like this. vol2 is all the .14, .10 and .2 spindles, vol1 is all the .13, .9 and .1 spindles. vol 4 is just the .12 and .8 spindles (smaller), vol0 (root volume) is 3 of the .0 spindles, and four of the .0 spindles are spares.
the bitch is that, when a drive fails and rebuilds to a spare, you have to re-rebuild the volume to get the raid-groups back into alingment.
Works for me, but YMMV.
Volume vol1 (online, normal) (block checksums) Plex /vol1/plex0 (online, normal, active) RAID group /vol1/plex0/rg0 (normal)
RAID Disk Device HA SHELF BAY CHAN Used (MB/blks) Phys (MB/blks) --------- ------------ --------------- ---- -------------- -------------- parity 9.15 9 0 11 136000/278528000 138959/284589376 data 7.15 7 0 11 136000/278528000 138959/284589376 data 6.15 6 0 11 136000/278528000 138959/284589376 data 10.15 10 0 11 136000/278528000 138959/284589376 data 8.15 8 0 11 136000/278528000 138959/284589376 data 3.15 3 0 11 136000/278528000 138959/284589376 data 2.15 2 0 11 136000/278528000 138959/284589376
RAID group /vol1/plex0/rg1 (normal)
RAID Disk Device HA SHELF BAY CHAN Used (MB/blks) Phys (MB/blks) --------- ------------ --------------- ---- -------------- -------------- parity 7.11 7 0 7 136000/278528000 138959/284589376 data 9.11 9 0 7 136000/278528000 138959/284589376 data 8.11 8 0 7 136000/278528000 138959/284589376 data 10.11 10 0 7 136000/278528000 138959/284589376 data 6.11 6 0 7 136000/278528000 138959/284589376 data 3.11 3 0 7 136000/278528000 138959/284589376 data 2.11 2 0 7 136000/278528000 138959/284589376
RAID group /vol1/plex0/rg2 (normal)
RAID Disk Device HA SHELF BAY CHAN Used (MB/blks) Phys (MB/blks) --------- ------------ --------------- ---- -------------- -------------- parity 7.3 7 0 3 136000/278528000 138959/284589376 data 9.3 9 0 3 136000/278528000 138959/284589376 data 8.3 8 0 3 136000/278528000 138959/284589376 data 10.3 10 0 3 136000/278528000 138959/284589376 data 6.3 6 0 3 136000/278528000 138959/284589376 data 3.3 3 0 3 136000/278528000 138959/284589376 data 2.3 2 0 3 136000/278528000 138959/284589376
(e.g.
For example, if I wanted to create a volume with 3 disks, would it be better to pick three disks on the same bridge (like 3.0, 3.1, and 3.2) or would it be better to pick three across three bridges (like 3.0, 3.8, and 3.12) ? Or doesn't it make a difference?
Thanks,
Mark