[Gluster-users] Performance question
Arnold Krille
arnold at arnoldarts.de
Sun Feb 12 23:37:10 UTC 2012
Hi all,
I know my hardware is not high-performance. But during my gluster-tests there
are some remarkable results...
My node are:
node1: Home-brewn home-server based on an atom.
node2: Old heater named Pentium4
node3: Turion64-based laptop
The first have their disks via sata and are connected via Gigabit-Ethernet, the
laptop has only pata and 100MBit.
I did run dbench with the same parameters but varying number of clients on the
raw disks of the bricks and also on the fuse-mounted volume in its different
stages of 1) two nodes, two bricks, 2) two nodes, four bricks and 3) three
nodes, 6bricks. The results of Throughput and Maximum Latency are graphed in
the attached pdf[*].
The third node was only used with the three-node-setup.
I find it remarkable that the local disk is faster by a factor of two in
throughput and faster by a factor of ten(!) in latency.
The first I could explain as the write-access has to go through the network and
to the disk and might thus get a performance-penalty on this old, single-core
system. But the high latency is very strange...
I didn't do any special optimization, mounted all the brick-partitions as ext4
with defaults,user_xattr,noatime and there was nothing else going on on the
cpus and network apart from normal desktop and background-music.
Did I do something wrong, did I miss some optimization or is this to be a
expected performance?
Thanks for your input,
Arnold
[*] I hope that makes it through to the list and doesn't offend because of its
size.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bigreplication.pdf
Type: application/pdf
Size: 45792 bytes
Desc: not available
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20120213/a50096be/attachment.pdf>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part.
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20120213/a50096be/attachment.sig>
More information about the Gluster-users
mailing list