Can anyone suggest what filesystems are suitible for scratch disks? note majority of io will be via mmap.

By scratch i mean scalable to many terabytes, but for short lived data that can be regenerated fairly easily (all be it with a cpu cost), so occasional disk failures resulting in data loss are acceptable, and actually preferable vs reduced read performance.

ZFS the solaris default takes an unacceptably long time to delete stuff, and is wasteful on memory under mmap io. tmpfs has 2 problems, no sparse file support, and i want to keep the scratch disk separate from /tmp since i dont want filling it up bringing the system down. /tmp needs to be mirrored anyway.

I'm almost thinking a ufs striped array might be best here. When I was running solaris 10 I never had the trouble deleting stuff that I do since upgrading the hardware. Any other thoughts?

  • /tmp needs to be mirrored anyway. That seems strange if you can accept your scratch device failing. – Andrew Henle Nov 17 '15 at 11:50
  • not really. if tmp fails the system comes down. if a scratch device fails, a process using it may crash - but not much else – camelccc Nov 17 '15 at 13:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.