Folks, I may have found the root cause, when I use a loopback "block device" both Dom0 and DomU cache the metadata, and this affects Venti's initialization somehow even if I increase the vfs_cache_pressure kernel tunnable. The answer here is to use some real block devices (no loop0) to avoid this caching issue and eventually, from time to time:
echo 3 > /proc/sys/vm/drop_caches to wipe the metadata cache if you are using a loopback device for the plan9 cdrom. Since I've moved to real block devices, I can not repro this issue, which is a satisfactory workaround. Thank you for looking into this, if one has the same problem with Xen, I strongly recommend to use gparted and create some block devices for plan 9 until linux is mature enough to disable metadata caching on loopback devices, which doesn't appear to be the case as of today. Jerome Ibanes On Mon, 26 Mar 2007, erik quanstrom wrote: > i've had this problem. it came from not shutting > down venti properly before reboot. venti is very > sensitive about this. i haven't used venti in quite > some time (kenfs is less sensitive) so i don't quite > remember the solution. but there are some venti > maintence commands that'll fix the problem. > > - erik > > On Mon Mar 26 09:02:25 EDT 2007, [EMAIL PROTECTED] wrote: > > > Task to do [fmtventi]:[enter] > > > Venti arena partitions to use [/dev/sd00/arenas]:[enter] > > > Venti index partitions to use [/dev/sd00/isect]:[enter] > > > > Did this step appear to work properly - no warnings or strange > > messages? > > > Jerome Ibanes <[EMAIL PROTECTED]> Infrastructure Architect & Systems Network Security Administrator PGP key fingerprint: 3FD68890 6D660FBE F92539EF 6F7BAC03 68151C51
