On 11 Jul 2012, John Mark Walker wrote:
> Further enhancements down the road will enable better performance for other use cases, but for now, don't expect a VM residing on a GlusterFS volume to have great write performance.
Just as a sanity check for our own setup, write
performance out of the box with KVM (managed via libvirt) is
awful, like less than 5MB/s because the disk cache mode is set to
none by default.
But changing this to writeback or writethrough is still
considered acceptable and safe, correct? This gets the
performance up to around 40-50MB/s. While this isn't stellar by
any means, it's still faster than our old NetApp via NFS.
The reason I ask is because the latest libvirt with RHEL
6.3 is now considering all of our defined VM's using writethrough
cache mode unsafe, so we have to live migrate using the --unsafe
flag. But my understanding from reading previous posts on this
list is that using something other than a cache mode of none is
acceptable and safe with Gluster at least.
Please correct me if I'm wrong.
--
Mark Nipper
nipsy at bitgnome.net (XMPP)
+1 979 575 3193
-
"If the fool would persist in his folly he would become wise."
-- one of the Proverbs of Hell from William Blake's _The
Marraige of Heaven and Hell_, 1789-1790
