On Fri, Dec 07, 2007, Amos Jeffries wrote: > > I've got too much on my plate to fix it at the moment, and it'd require > > a COSS format change. > > I don't think so, just a stable default of /var/cache/* or something for > COSS type dirs for a quick-fix. Re-format later as COSS-2 if its > sponsored. Nah the trouble is the rebuild times. Having to rebuild by reading the whole disk in megabyte chunks is just horribly inefficient. Assuming writeback on the disk is disabled and the disk Io is implemented correctly we could "mostly assume" that if we write the metadata out -after- we've successfully written out the stripe object data then the data will be valid; then we just need to read the metadata areas at the end of each stripe (which would be on the order of kilobytes, not megabytes) and the rebuild times would be a bit faster than a disk fsck. adrian -- - Xenion - http://www.xenion.com.au/ - VPS Hosting - Commercial Squid Support -
