[Ocfs2-users] 6 node cluster with unexplained reboots

[Alexei_Roudnev]

Mark, you are wrong. You CAN (and even SHOULD) use controller in write-back 
mode with the cluster -
any such controller have a power-independent memory and syncronize cache 
between cluster nodes (if SAN itself is a cluster).
Running SAN systems in Write-thru mode degrade performance dramatically 
(sometimes 2 - 10 times) and can be recommended
only for super-mission-critical tasks (because such systems have a danger of 
losing inromation, if they lost power for more then a few days so that 
batteries can't hold cache long enough - typical time is 72 hours).

Cash works AFTER all Linux nodes connects to the controller, so it is 
transparent for the controller (but it can create a jitter in writing time, 
true).

> On Wed, Aug 15, 2007 at 05:43:14PM -0700, Ulf Zimmermann wrote:
>> The SAN is a 3Par E200, which does write into cache on its two
>> controllers, then acknowledges a write and then writes it actually to
>> disk. I have not found any reason for this delay yet, so sofar I am
>> stumped why it had such a long delay writing.
>
> Are you saying that the controllers are doing write-back caching? If 
> they're
> in that sort of mode, you need to change it to write-through for a 
> clustered
> environment.
> --Mark
>
> --
> Mark Fasheh
> Senior Software Developer, Oracle
> mark.fasheh at oracle.com
>
> _______________________________________________
> Ocfs2-users mailing list
> Ocfs2-users at oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>