Ben Browning wrote:
> Eric Shubert wrote:
>> Nice piece, Ben. I might add that with a reduction in hardware comes
>> an increase in reliability.
>
> To some degree, anyway... In N+1 clustering solutions, more hardware
> leads to better stability.
Good point. My thinking context was in the absence of clustering and
raid. These are must-use technologies for stability.
> I once had two servers with mission-critical
> services on them both throw hard drive fits one night(they had 2x scsi
> drives, but not RAIDED as we were using one exclusively for mail queue
> IO), so I simply limped them along long enough to drain their queues and
> halted them, dealing with them the next afternoon...
>
Interesting.
I heard a story of a system with 2 raid1 drives. The drives came from
the same lot which had a manufacturing defect. They died at the same
time! :(
>> I worked on a server some time ago that had SCSI drives which had a
>> MTBF of 36 years. The server had 72 of them. One failed every 6
>> months, like clockwork.
>
> I had a RAID under my control that had not been powered down in 5 years.
> When we finally did, half the drives did not spin back up :)
Amazing what a little static friction can do!