Hello,
> Hi, I'm researching possible cluster solutions for a web service that serves
> lots of relatively small files (2MB or less on average).
We have quite a few users who are using glusterfs as a web backend
storage. Though we had not thought of this community initially,
glusterfs seems to work well, or rather, this community has a greater
requirement for such a ssolution.
> I've been leaning toward mogileFS because I know it's deployed on a large
> production system (LJ) doing exactly what I need. It also provides replication
> and fault tolerance.
> I just found out about glusterfs, and am very impressed with the clean design
> and the fact that it works at the posix fs level so I won't need to change my
> app.
>
> Is glusterfs deployed in any large production system?
Yes, we have glusterfs running in a few production sites.
> I read through the Automatic File Replication examples on the web site, and
> while I understand how they work, it seems like a lot of set up is required,
> especially as you add more nodes. Is there a way to configure glusterfs to
> work more like mogilefs, where a file is saved to 2 or 3 of the storage
> servers automatically? It seems like the RR scheduler almost does what I want,
> if it could just save the file to a couple of more subvolumes instead of just
> one.
this is an idea which we have evaluated before (unify+scheduler doing
the 'mirroring' as well). but for more than one reason decided unify deals
with 'single copies' of files and AFR abstracts multiple copies into a
single copy. yes, you have to declare a few more volumes in the spec
file, which we think is easily scriptable (to generate the spec files
for large clusters with AFR). And ofcourse you can always get our help
at the mailing list or irc if you are having issues related to the
spec files :)
avati
--
ultimate_answer_t
deep_thought (void)
{
sleep (years2secs (7500000));
return 42;
}
