[OpenAFS] Optimizations
Lars Schimmer
l.schimmer@cgv.tugraz.at
Tue, 23 Apr 2013 11:06:02 +0200
This is an OpenPGP/MIME signed message (RFC 4880 and 3156)
------enig2ICMAIMVGQJNOQEKBLBDB
Content-Type: text/plain; charset=ISO-8859-15
Content-Transfer-Encoding: quoted-printable
On 2013-04-23 10:14, ProbaNet - Info wrote:
> Hello guys!
> We're using openafs 1.6 fileserversand we would like to know if we
> can do something to improve performance (we mean: parameters and values=
).
> Scenario:
> - 3 fileservers, all 64 bit KVM virtual machines with4 vCPU, 1280 MB RA=
M
> and 4 x 250 GB vice partitions (which are iSCSI targets)
One note: thats quite low memory. Our Fileservers (real hardware) do
have 16GB for linux caches on the FS. That does help. And we do have
iSCSI also - which does need some space for cache, too.
With this setup you need to check the iSCSI connection and the storage
itself - e.g. fine to have 3 FS, but what if the storage is a breaker in
this case, 2 RW volumes from 2 FS on 2 iSCSI targets on the same RAID6
is e.g. a breaker...
Also all 3 FS on one KVM Host (physical machine) is not a wise idea,
performance wise...
> - 2 dbservers (1 sync site + 1 non-voting clone), 64 bit KVM vms, 2
> vCPU, 512 MB RAM
Non-voting-clone? Usual both servers do sync and both are available to
clients. Those do not need much.
> - everything runs into an office LAN (1 Gbit), with around 40 users
> - each fileserver hosts 200 to 600 volumes
Yeah, quite usual.
> Here's actual BosConfig for dbservers:
> -----
> restarttime 16 0 0 0 0
> checkbintime 16 0 0 0 0
> bnode simple ptserver 1
> parm /usr/lib/openafs/ptserver -p 16
> end
> bnode simple vlserver 1
> parm /usr/lib/openafs/vlserver -p 16
> end
> -----
Not much to say, dbservers are quite low profile, but important ;-)
> Here's actual BosConfig for fileservers:
> -----
> restrictmode 0
> restarttime 16 0 0 0 0
> checkbintime 16 0 0 0 0
> bnode fs fs 1
> parm /usr/lib/openafs/fileserver -L -udpsize 65536 -cb 600000 -rxpck
> 4000 -vattachpar 8 -realm XXX.YYY
> parm /usr/lib/openafs/volserver -p 16 -udpsize 65536
> parm /usr/lib/openafs/salvager -parallel all4 -orphans remove
> end
> -----
Ours (after some discussion in ML):
parm /usr/lib/openafs/fileserver -L -udpsize 131071 -sendsize 131071
-rxpck 700 -p 128 -b 600 -nojumbo -cb 1500000
parm /usr/lib/openafs/volserver -nojumbo -udpsize 131071
which does start some processes in parallel (hence the memory).
> Everything works quite fine (except for some occasional slowdown when
> accessing some volume for the first time), but we would like to know if=
> we can do something to improve it! :) Thank you very much!
Depends on the current values you got currently ;-)
Our setup is nice in the 30-50 MB/sec usage in every-day use.
Up to 90 MB/sec on big files (images) read from/to cache (memory...).
> Stefano
> Fabio
MfG,
Lars Schimmer
--=20
-------------------------------------------------------------
TU Graz, Institut f=FCr ComputerGraphik & WissensVisualisierung
Tel: +43 316 873-5405 E-Mail: l.schimmer@cgv.tugraz.at
Fax: +43 316 873-5402 PGP-Key-ID: 0x4A9B1723
------enig2ICMAIMVGQJNOQEKBLBDB
Content-Type: application/pgp-signature; name="signature.asc"
Content-Description: OpenPGP digital signature
Content-Disposition: attachment; filename="signature.asc"
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)
iEYEARECAAYFAlF2TvoACgkQmWhuE0qbFyMWMgCfYncA3fpffIYLEA0QjKAf67w+
rksAniAr5uplioU6UVmg1KNTHCV6MTj/
=4FOh
-----END PGP SIGNATURE-----
------enig2ICMAIMVGQJNOQEKBLBDB--