I can’t even blame it on reading xkcd before bed, because I wasn’t this time.
But something needs to change.
Tonight, again, I dreamed about gatecrashing Donald Knuth’s secret school for hackers. As part of my initiation I threw a bottle of rum at a bunsen-burner equipped car.
My subconscious is definitely trying to tell me something. Unfortunately, the message is “Sam, you’re a psycho.”
Client’s backup server died. It was brought up in .. looks like May of 2005, with the main backup target being a RAID5 set consisting of seven 80GB partitions distributed over six disks. The RAID members were (hda3, hdd1, hde1, hde2, hdf1, hdg1, hdh1). There was also a RAID1 set (hda1, hdc1) of about 30GB. The CD was hdb.
Those of you keeping score at home will note that having seven RAID members but only six disks means that two RAID partitions reside on the same disk. Two partitions on the same disks means that a single disk failure knocks out two partitions, and RAID5 is only robust against a single failure. So originally the RAID set had five active disks and two pre-configured spares, with one of the two partitions on hde (the 160GB drive) active and the other set to spare.
That worked for a very short while until the IDE card we’d had to install (to get the extra 4 IDE devices) started consistently failing to give us hdg and hdh. After a while of failing to get this to work, I gave up and let the RAID5 run with only five partitions on four devices - (hda3, hdd1, hde1, hde2, hdf1). This was the configuration it had until Wednesday, when it died.
So the RAID5 set, when running, yielded 294GiB of redundant storage, from a base of about 374GiB. (Some disks were slightly larger than the 75GiB size of the smallest contributing disk, and RAID5 always uses the smallest size disk to construct its parity sets.)
You can see what the problem with this server was, I think. Its capacity was grotesquely out of date. Today I can buy a new 500GB disk from Memory Express for $120.