Opened 9 years ago

Last modified 4 years ago

#255 new enhancement

AFS cache partition

Reported by: ezyang Owned by:
Priority: major Milestone:
Component: internals Keywords:
Cc:

Description (last modified by ezyang)

This seems to work reasonably (no journaling since we don't care about crash recovery, needs lots of inodes). 11G, since AFS checks for at least 5% extra room for our 10G cache.

  mkfs.ext4 -O ^has_journal -m 0 -N 1000000 /dev/xvds1

Need to migrate s-a and p-b, and all of the new F15 servers. Also update the Kickstart.

Given mild testing seems to have fixed major load problems, but we should wait for another 24hr cycle to be sure.

If corruption is detected, we don't want to fsck. Instead, blow the partition away and make a new filesystem. This requires us to do some boot script hacking.

Change History (6)

comment:1 Changed 9 years ago by andersk

  • Description modified (diff)

comment:2 Changed 9 years ago by ezyang

  • Description modified (diff)

comment:3 Changed 9 years ago by ezyang

  • Milestone changed from Fedora 15 to Fedora 17
  • Priority changed from major to minor

It was a benefit, but not major enough. Downgrading importance and bumping milestone. We can do this when we feel like it.

comment:4 Changed 9 years ago by ezyang

  • Milestone Fedora 17 deleted

comment:5 Changed 9 years ago by ezyang

  • Type changed from defect to enhancement

comment:6 Changed 4 years ago by andersk

  • Priority changed from minor to major

AFS reacts very poorly to the cache partition becoming full; this seems to cause user-visible file corruption. So we should do this independently of any performance concerns.

Note: See TracTickets for help on using tickets.