On 05/01/2013 01:50 PM, Jay Vyas wrote:> There has been chatter about "X on gluster", where x=mongo,
riak,...
>
> Im wondering - is there a "most popular" or most well tested
> transactional datastore that runs/leverages gluster ?
>
> Or is the idea of running a transactional nosql tool on gluster still
> mostly a fun/cool/interesting thought experiment?
I follow developments in the NoSQL world pretty closely, and count many
people in that space as my friends. This idea comes up often, but
nobody really pursues it much because what they do and what we do is
already so similar. The consistent hashing we use in DHT is clearly of
the same general sort as that used in Cassandra, Riak, or Voldemort.
Some of the discussions we've had about various forms of replication and
different consistency models clearly relate well to those same concepts
in MongoDB or Couchbase. If we're using the same algorithms for things
like distribution and replication already, why put one on top of the
other? Putting Cassandra on top of GlusterFS would be too much like
putting Cassandra on top of itself.
That said, there are a couple of related ideas that are somewhat
interesting. Most have to do with splicing pieces of these related
technologies together instead of layering them. What if we could layer
our front end (full POSIX via FUSE plus SMB/NFS support) on top of their
back end? What if we could put their API on top of our back end with a
specialized translator or libgfapi, much as we're doing for Swift and
Hadoop? There are plenty of possibilities like that to explore.