Messages in this thread | | | Date | Fri, 25 Jan 2008 12:37:05 -0500 | From | "Brett Dikeman" <> | Subject | Re: 2850 drive bays not hot-swap? |
| |
On Jan 24, 2008 3:35 PM, William Warren <hescominsoon@emmanuelcomputerconsulting.com> wrote:
> http://lists.us.dell.com/pipermail/linux-poweredge/2003-July/008898.html
An update: I don't know whether it is a kernel problem or a hardware problem, but the instructions pertaining to removing and adding SCSI devices live above doesn't work on a PE2850 (which uses "LSI Logic / Symbios Logic 53c1030 PCI-X Fusion-MPT Dual Ultra320 SCSI (rev 08)", aka Dell's PERC controller 4e/Di, driven by the mpt* drivers.)
It *appears* to work, but the bus seems to grind to a near halt. This may be an inaccurate conclusion, but I'm leaning toward kernel issues, since iowait didn't skyrocket.
ID's 2-5 on a U320 bus off a PERC controller are in a RAID10 array. IDs 0&1 are in a separate mirror.
1)echoed "scsi remove-single-device 0 0 5 0" > /proc/scsi/scsi
2)Pull the drive and issued mdadm /dev/md0 --remove detached
4)Re-inserted the same drive and issued echo "scsi add-single-device 0 0 5 0" > /proc/scsi/scsi (and the kernel gave it /dev/sdg instead of /dev/sdf, as a side-note.)
5)issued mdadm /dev/md0 --re-add /dev/sdg1
No messages indicating anything afoul.
I then fired up watch -n 5 /proc/mdstat and noticed that the rebuild rate was a fraction of what it should have been; hovering at a very constant number- I believe 2339KB/sec. It should be 74,000KB/sec (drops to as low as 45,000KB/sec at the tail end of the drives.) Load average jumped to 2+, yet nothing showed up in top as taking any CPU time, even with a 10 second average. No iowait indicated. I increased sync_speed_min to 20000 with no change.
Then I noticed that the system drive mirror (first two drives) looked to have quite a bit of disk activity, but I couldn't find what was causing it. Now I realize that it wasn't "quite a bit"- I think it was disk I/O grinding to a near halt.
I shut down the system, and after power-up and booting, the RAID10 array is rebuilding at full speed...
I'm happy to provide additional info and try things for interested parties to troubleshoot or fix this problem. Being able to hot-swap drives is pretty critical here.
Brett
| |