What I love about GlusterFS more than anything, is it's ease of use to
setup, especially on the client side. Install the binary and you're
done. Scaling the storage is also relatively convenient (even more so
with upcomming the 1.4 release). But sometimes you don't need to scale
storage as much as you just need to scale to handle a large volume of
requests. In such a case, it's unnecessary to disrupt the cluster to add
in another node, especially if you will just bring that node down in a
few hours anyways. In such a case, it would be more convenient to add in
caching nodes.
My random thought is to introduce a Memcache translator (not sure if
that's the correct term) that would work along the lines of CacheFS.
Writes would be mirrored like in AFR, once to the cluster another to the
Memcache pool. Clients could attempt to read first from the Memcache
servers or based on some formula, e.g 65% of the time. The size of files
stored would be configurable, e.g. only if less than 50K bytes. Stat
caching could also be extended to utilize Memcache as well.
What are your thoughts on this?
Best,
Erik Osterman
