Ocfs2 users - Sep 2009 - Two node performance question

Hi,
I'm new to OCFS2 and have a question about my setup. I have two SLES 10
nodes (no patches yet) and a single OCFS2 volume. The volume receives live data
via replication to keep it up to date. The goal is to have node one receive
replicated date and all nodes run reports off this date. Today I am just using
two nodes to get started.

The shared storage is an Openfiler 1.3 system running iscsi. Our numbers for raw
read and write look good.

The problem is I am having an issue with keeping both nodes fast. If node 1 is
running reports quick, then node two is very bad and vise versa.

I've run hdparm -t /device on both nodes at the same time to see if I get a
drop off and it certainly does.

Also, the bulk of the data is being written into /folder and all nodes are
reading out of that same folder at the same time. The files are basically a slat
file database that records are written into via one node and read from all to be
more specific. So a lot of SMALL writes (one node) and many SMALL read (all
nodes).

I have no clue how to optimize this or if that is possible, if one version will
handle this better then another. We're on base SLES 10 and I didn't want
to buy patches until I had some guidance.

Thanks in advance

Bruce A Leggett
Director, IT Systems





This Email message and any attachments are confidential.  If you are not the
intended recipient, please notify Amscot Financial, Inc. immediately by replying
to this message and destroy all copies of this message and any attachments.Thank
You.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://oss.oracle.com/pipermail/ocfs2-users/attachments/20090918/f56ffd32/attachment.html

Bruce Leggett wrote:
>
> Hi,
>
> I?m new to OCFS2 and have a question about my setup. I have two SLES 
> 10 nodes (no patches yet) and a single OCFS2 volume. The volume 
> receives live data via replication to keep it up to date. The goal is 
> to have node one receive replicated date and all nodes run reports off 
> this date. Today I am just using two nodes to get started.
>
> The shared storage is an Openfiler 1.3 system running iscsi. Our 
> numbers for raw read and write look good.
>
> The problem is I am having an issue with keeping both nodes fast. If 
> node 1 is running reports quick, then node two is very bad and vise 
> versa.
>
> I?ve run hdparm ?t /device on both nodes at the same time to see if I 
> get a drop off and it certainly does.
>
> Also, the bulk of the data is being written into /folder and all nodes 
> are reading out of that same folder at the same time. The files are 
> basically a slat file database that records are written into via one 
> node and read from all to be more specific. So a lot of SMALL writes 
> (one node) and many SMALL read (all nodes).
>
> I have no clue how to optimize this or if that is possible, if one 
> version will handle this better then another. We?re on base SLES 10 
> and I didn?t want to buy patches until I had some guidance.
>
If concurrent io to a shared device from multiple nodes is slow, then
the fs performance will also be slow. This will be true for any workload
and any file system. So ignore your workload for the time being and instead
work on improving the concurrent io thruput of your iscsi setup.