Saltar al contenido ForoHackers

RAID0 for speed: a short post-mortem, with arithmetic

Copias de seguridad y recuperaciónHomelabSistemas de almacenamiento
#1091 · raid0_believer · · Denunciar

New here, and my first post is a confession. For two years I ran my home server on two drives striped in RAID0, because "the benchmark numbers were great". Last week one drive stopped responding. Both drives' worth of data went with it.

The arithmetic I should have done. Say each drive has a 3% chance of dying in a given year (a made-up but plausible number). With RAID0 the array dies if either one does:

P(array survives a year) = 0.97 * 0.97 = 0.941
P(array fails in a year) = 1 - 0.941   = about 5.9%

Striping roughly doubles the failure rate, because the array is only as reliable as the weakest of its parts in series. I traded reliability for speed, and I did not even have a speed requirement.

What I lost: a few hundred gigabytes of stuff I told myself was "replaceable". Some of it was. Some of it was photos.

What I am doing now: a mirror for availability, plus an actual backup on a different machine, and a calendar reminder to test a restore. I have read the old line "RAID is not a backup" about four times since last Tuesday.

#1713 · cron_job_carl · · Denunciar

Thanks for writing this down. Two small corrections to a good plan: a mirror does not protect you from deleting a file or from corruption (both disks get the same bad write), and "a backup on a different machine" should ideally also be a different place. And yes: test the restore, by actually restoring into an empty directory and comparing checksums.

#1714 · floppy_disk_farmer · · Denunciar

Sorry about the photos. If the dead drive is still detected at all, stop powering it on repeatedly and make a sector-level image of it with something that tolerates read errors (ddrescue is the one people trust), then work on the image, not on the drive. Every power cycle on a failing disk is a lottery ticket you do not want.

#1715 · m4gn3t1c_t4p3 · · Denunciar

Tape people have a saying: "backups are for restoring". An unread backup is a hope. You are already further than most, because you did the arithmetic before the second failure, not after.

#1716 · git_blame_myself · · Denunciar

For the record, I lost a week of work once by running git clean -fdx in the wrong directory. We have all been the cautionary tale at least once. The difference is writing it up.

#1717 · raid0_believer · · Denunciar

ddrescue is running now. It has recovered 91% so far and is slowly working through the bad sectors. If I get the photos back, there will be a very long post about 3-2-1.

#1718 · yak_shaving_daily · · Denunciar

A warning from experience: this will turn into a three-day project to set up the perfect backup system. Resist. Start with a plain scheduled copy to one other machine this weekend. Perfect is the enemy of restored.

Inicia sesión o únete para responder.