[OpenAFS] backup solutions

ted creedon tcreedon@easystreet.com
Sat, 19 Feb 2005 08:49:51 -0800


There is about 200GB on my 2 AFS servers about 12% of your 1.3 TB. However
each system has 500GB of capacity.

Since tape backups are a nuisance for me (but required for you - because its
not YOUR data) the setup here is to use 2 older 733MHz boxes on separate MGE
3KVA UPS's with HighPoint ide raid cards, 4 drives per card. Suse 8.x (2.4
kernel) runs on a separate drive.

/usr/vice and /vicepa are both on the raid drives so an entire raid
subassembly can be moved to a new computer if required.

I'm in the process of adding a 3rd box in a physically separate location. On
a 2KVA UPS.

The systems are protected from fire and (frequent) utility disturbances and
run 24/7 indefinitely without rebooting.

The reason for multiple servers is the restore time and on line redundancy
when (not if) a server fails.

Last week a drive failed and took the power supply with it. Most likely the
red phosphorous problem that all drives older than a year have. (I.e. all
disk drives have a congenital failure mechanism). It was a little scary
watching the 0,1 raid system rebuild itself but everything is back up and
running with no data outages.

I'd suggest 3 identical systems in separate physical locations using Dells
or better. (The motherboards are designed by Dell engineers, not some
vendor). I have write caching turned off but a modern scsi raid system won't
and it would help response time.

Both Morgan Stanley and Stanford use AFS so its ruggedness is a proven fact.
Stanford has petabyte level storage and 400GB/day new data from its Linear
Accelerator distributed worldwide. The setup at Stanford supports all the
class load and is cleverly done with minimum admin time. I'd recommend a
visit or a call.

Tedc