[Ocfs2-users] Performance degrade when more nodes adding

[Sunil Mushran]

How many nodes are you talking about?

Remember that file system/cluster software is not the only component
that is involved in scaling.

You have to consider the hardware involved too. If the storage array
is close to saturated with the number of nodes you have, adding another
node will not help. Same goes for the interconnect. Also, the cpu/memory
in each node. User mostly look at the memory requirement of the userspace
app ignoring the fact that the fs needs to cache the inodes and locks.
If the memory is low, the inodes will end up being freed forcing the fs/dlm
to remaster the same locks over and over again. Cpu also plays a role.
The lock synchronization takes some cpu. Meaning one can scale a dual
core node to only so many nodes. And that number depends heavily on
the workload.

Test your workload. Locate the bottleneck. If it is a software bottleneck,
then post your findings and we can include that in our todo list.

On 02/03/2011 11:24 PM, Thomas Lau wrote:
> Hi, according to cluster file system logic and from my testing experience on other FS, the more nodes add into clustered file system, the more performance degradation it will have, two questions:
> 1. Is it possible to tune ocfs2 so that the performance issue wouldn't be a breakdown problem?
> 2. How many cluster nodes do you guys experience will have major performance issue?
>
> The fs we use is 90% read, 10% write, mostly video which are more than 5MB, and small php files.
>
>
> "Sent via CSL BlackBerry."
>
> _______________________________________________
> Ocfs2-users mailing list
> Ocfs2-users at oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs2-users