[OpenAFS] Re: Recommended options for dafs in openafs 1.6.1?
Mon, 30 Apr 2012 17:09:37 -0500
On Sat, 28 Apr 2012 08:29:10 +0200
Staffan Hämälä <firstname.lastname@example.org> wrote:
> I'm in the process of upgrading one of our AFS servers to 1.6.1 (we've
> only run 1.4 before). I've been looking for information about
> recommended options for the dafs server, but I haven't found anything
> specific to 1.6 or dafs.
The pre-DAFS options all work the same as for DAFS (iirc the only thing
we removed was -Dontsalvage for the salvager). The only thing different
with DAFS is that you get new options; but most of them I think are a
"only touch if you know what you're doing" kinda thing. A couple you
might want to change may be...
This changes the number of hash buckets for volumes. If you start to get
like tens of thousands or hundreds of thousands of volumes on a server,
you may want to raise it.
This changes how often we detach an idle volume (that is, how many
seconds of inactivity are required before we detach it). If you have
specific knowledge of the access patterns for volumes on a server, you
may get better performance by changing this, and maybe some of the other
-vlru* options. But for most uses, it doesn't matter.
> Are there any recommendations?
> We use the following options on our 1.4 fileservers:
> /usr/afs/bin/fileserver -p 128 -b 100000 -l 10000 -s 10000 -vc 100000
> -cb 200000 -rxpck 10000
'-b 100000' seems pretty huge. The hash table size for dir buffers is
only 32, so assuming even distribution in the hash table, you've got
hash chains thousands of entries long. I'm not sure if that's a good
idea, but it depends on access patterns.
'-rxpck 10000' shouldn't be necessary anymore. This depends on the
versions in play, but certainly not for 1.6.
'-vattachpar N' might help you depending on your partition layout and
number of volumes, but if you're not bothered by your startup/shutdown
time, it's not going to make any difference.