Hi Alexander.
On Tue, Dec 08, 2009 at 08:12:13PM +0300, Alexander Beregalov
([email protected]) wrote:
> I have tried latest server from the git (Added dump/del all indexes
> aka config groups commands)
Good to know!
> with 2.6.31 plus 2d7cf8ef75 (Staging: pohmelfs: sync with the
> development tree) and 2.6.32 with my fix of NULL pointer dereference
> (which seems is not correct) as a client. Both with the same result.
>
> My testcase:
> tar jxf linux-2.6.31.tar.bz2
> make defconfig
> make -j8
>
> Server (UP) exports tmpfs, client (4 cores) runs on another host, 1GbE
> connection.
>
> Untaring takes 90-100 seconds. At the beginning there is no load on
> both hosts (only 15kb/s rx and tx), then server goes to 100% cpu load
> in system and soft irq, network load is 3-6Mb/s at that time - it runs
> in that way for few seconds, than again becomes idle.
That's the way local cache works - client may decide to flush to
server way after it was written, and generally it will be batched into
rahter large portions, which may explain above behaviour.
> Some time after tar is completed I see that uid\gid ownership fall to
> 0:0 (root).
Yup, POHMELFS in its current incarnation does not support proper user
IDs, since there is no metadata objects.
> And some files I cannot read - No such file. It can be read on the
> server, but client can't.
> I try to run "make defconfig" - sometimes it runs, somtimes it does
> not (permission denied, no such file)
That looks like a bug. If the fact that POHMELFS is being under active
switch to completely different storage system does not affect you,
kernel debug output and information about where system failed, i.e.
which files were not found and so one, will help.
Thank you!
--
Evgeniy Polyakov
_______________________________________________
Pohmelfs mailing list
[email protected]
http://www.ioremap.net/cgi-bin/mailman/listinfo/pohmelfs
