deadlock with 2.6.9
Quoted 9 lines · On Sun, 07 Nov 2004 at 09:22:14 +0100 Bernd Eckenfels wrote:
Quoted 2 lines
Get a real RAID controller (3Ware, not some crappy pseudo-RAID junk.) They are much more reliable than software RAID.
On what sample do you base this claim?
Generally hardware raid sooner or later makes problems (especially if you run raid5 in degenerate mode or try to rebuild by disk replacing with differen/old signature). Also bus hangs are commonly not very well received by hw raid firmware or drivers.
I had 28 mirror sets on Compaq SMART2/p controllers in one server (four
controllers, two SCSI channels each, seven disks per channel.) All the disks on channel A of each controller were mirrored to those on channel B, so even complete failure of one channel didn't cause a problem.
Once a disk was marked 'failed' in the controller NVRAM there was no way to
convince it that some newly-inserted disk contained valid data.
Booting up with SCSI cables connected wrongly (channels A and B swapped) got you
a nice error message informing you of this fact. Swapping SCSI IDs on different disks on the same channel was also detected and reported nicely.
And attempting to boot with a bad cable (bent pin) gave a message saying 'either
power down NOW and check cables or I will mark every disk on that channel as failed.'
Of course this system was 100% Compaq; even the disks had Compaq firmware
though the labels said IBM. And it was very expensive...
--Chuck Ebbert 12-Nov-04 22:42:32
Quoted 26 lines · Chuck Ebbert
Quoted 9 lines · On Sun, 07 Nov 2004 at 09:22:14 +0100 Bernd Eckenfels wrote:
Quoted 2 lines
Get a real RAID controller (3Ware, not some crappy pseudo-RAID junk.) They are much more reliable than software RAID.
On what sample do you base this claim?
Generally hardware raid sooner or later makes problems (especially if you run raid5 in degenerate mode or try to rebuild by disk replacing with differen/old signature). Also bus hangs are commonly not very well received by hw raid firmware or drivers.
I had 28 mirror sets on Compaq SMART2/p controllers in one server (four
controllers, two SCSI channels each, seven disks per channel.) All the disks on channel A of each controller were mirrored to those on channel B, so even complete failure of one channel didn't cause a problem.Once a disk was marked 'failed' in the controller NVRAM there was no way to
convince it that some newly-inserted disk contained valid data.Booting up with SCSI cables connected wrongly (channels A and B swapped) got you
a nice error message informing you of this fact. Swapping SCSI IDs on different disks on the same channel was also detected and reported nicely.
Quoted 6 lines
And attempting to boot with a bad cable (bent pin) gave a message saying 'either
power down NOW and check cables or I will mark every disk on that channel as failed.'Of course this system was 100% Compaq; even the disks had Compaq firmware
though the labels said IBM. And it was very expensive...
I have had quite good luck with the IBM "ServeRAID" controllers, running 30+ systems with one or two of them and between 1-10TB storage. To convince the controller that a drive marked bad really is good does require shutting down the system and saying reassuring things to the firmware from the console, but it can be done. I'm told that I may be able to do that running now, using the control application, but given that the servers are spread over four timezones I have no urge to try the remote version.
I would like to have some form of hardware RAID on at least what I need to boot, so a failure of one of the boot drives doesn't leave the machine unable to boot. Particularly with (S)ATA systems which tend to fail to the mirror only when the first drive is *dead*, not when it's too sick to give good data, but still trying.
Signature · 3 lines
-bill davidsen ([email protected]) "The secret to procrastination is to put things off until the last possible moment - but no longer" -me