[PVFS2-developers] more flow/job timeout comments?

[Nathan Poznick]

Thus spake Phil Carns:
> I don't really know what's going on, but from combining comments from Pete, 
> RobL, and Neill, it seems like there is a pattern of the server disks getting 
> too busy under load and causing some flows to periodically stall while the 
> server tries to flush to disk all at once.  We still need to look at it 
> closer to know for sure.

I mentioned this on IRC, but I know that mythtv (a homebrew PVR system)
periodically (every few seconds) fsync()'s its output files while
recording.  It does this to avoid filling up several hundred megabytes
of memory with VFS cached pages and then triggering a stall as the
kernel attempts to free up the older cache by dumping to physical disk.

Are the stalls happening when only using async, or when using sync as
well?  If they only happen with async, a similar situation may be
occuring, and it may be wise to from time to time sync explicitly, even
when using async mode.

-- 
Nathan Poznick <poznick at conwaycorp.net>

Language is the armory of the human mind, and at once contains the
trophies of its past and the weapons of its future conquests. - Samuel
Taylor Coleridge

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://www.beowulf-underground.org/pipermail/pvfs2-developers/attachments/20040716/8fb7e55f/attachment.bin