On 12/20/11 5:39 PM, Gregory Farnum wrote:
On Tue, Dec 20, 2011 at 3:30 PM, Matt Weil<mweil@xxxxxxxxxxxxxxxx> wrote:
Hello all:
given six osd servers does it make any difference of each one also runs a
mds or not?
How many mds's is to many. How does this effect metadata performance.
Right now a single-MDS system is significantly more stable than a
system with multiple active MDSes, so if possible you should use that
configuration. Only add MDSes when you need more metadata throughput
(probably you only need one).
Thanks for the response.
these osd servers are also going run the client and serve out NFS.
Is it better to split the mds and mon out to a dedicated system? then
run standby on the active osd's
Adding more MDSes should increase metadata throughput; it wouldn't
surprise me if you could construct a set of operations that slow down
but you're not going to run into one by mistake.
However, if you just start up a metadata server on each OSD node
without changing the max_mds parameter, what you'll actually get is
one active MDS and 5 standby daemons, which will be fine and give you
lots of changes to fail over your MDS in the event of crashes. ;)
-Greg
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
