[OpenAFS] openafs performance in an internet-scale traffic environment
Wed, 21 Feb 2007 14:46:45 -0500
a couple of things to think about before plowing ahead...first, don't
forget about the afs cache on your webservers, and think about how you
will update files and coordinate the vos releasing of the volumes.
if it is critical for the webservers to all see exactly the same data at
exactly the same time, and if there is frequent updating of files, you
really have to separate out the active stuff from the quiescent stuff.
that way, vos releases should only minimally lock things up.
i worked as a consultant on an afs-backed website at a financial place a
few years back. they had huge volumes, vos releases triggered
automatically each time a file changed, webservers and afs servers on
different continents....and, not surprisingly, the perceived
performance was bad.
i also know that afs-backed webservers run great in terms of storage and
allowing users to "publish" data. but you may have to do some tinkering
of the webserver or the afs client, and you definitely want to have a
good idea about the volume layout and replication strategy before you begin.
Richard Human wrote:
> Yes, although I haven't quite got my head around how I would fit the
> replicated volumes into the architecture. I was thinking of a scenario
> where I would have the r/w volume on one server in the cell and would
> replicate to at least 1 other server in the cell. So probably at least
> 3 servers in the cell. (Can you mix replicated r/o volumes and r/w
> volumes on one server?)
> This fits with our application as uploads don't happen nearly as often
> and the source is an application server on the back-end. So we can
> push the files onto the r/w volume and then let them replicate out to
> the r/o volumes. We can even afford a delay of minutes here.
> Does this make sense w.r.t. to AFS?
> Anyway, I still need to go through a test setup, I'm just looking for
> a sanity check that this type of scale is possible with AFS and that
> I'm not trying something stupid.
> Thanks for your reply
> On 21 Feb 2007, at 8:21 PM, Christopher D. Clausen wrote:
>> Richard Human <email@example.com> wrote:
>>> So we're looking at open-afs. We are thinking of a situation where we
>>> have a cell of 2+ servers on the back-end with our data spread over a
>>> number of volumes, that we can move around. This takes care of the
>>> scalability issue. For redundancy we would investigate the volume
>>> replication feature.
>> You are aware that AFS replication is for read-only data? You MUST
>> write to the read-write volume and then vos release the data to the RO
>> OpenAFS-info mailing list
> OpenAFS-info mailing list