Ocfs2 users - Nov 2005 - Heartbeat threshold, misscount & hangcheck co-ordination

Hi,

Having configured 2-node RAC on OCFS2 I'm trying to establish the best
figure for the heartbeat threshold. I've seen a few threads on this, but
they don't address the other timeout values in the system.When configuring
OCFS 1, I didn't notice a heartbeat threshold: either I missed it completely
or it is a new function.

With the heartbeat default at 7 (2 second ticks) that puts a self-fencing
situation at (threshold -1) *2  = 12 seconds.
However CRS has a misscount parameter default of 60 (crsctl get css misscount),
which is the time that a node will take before evicting another node from the
cluster when it fails to respond across the interconnect.

Additionally, the hangcheck-timer module must be co-ordinated with this
misscount parameter to ensure that if a hung node revives itself after being
evicted from the cluster, it reboots to avoid corrupting the database during an
attempt to resume previous transactions.

It strikes me that the heartbeat threshold should probably also be co-ordinated
with these other 2 parameters: ie, if a node self-fences then the rest of the
cluster should be evicting it at the same time, not waiting a further 45
seconds.

While all the Oracle documentation for 9i RAC and CM on OCFS (1) did not
recommend a lower value than 60 seconds for misscount, I found that a bit high
in terms of cluster timeouts for a TAF scenario, especially as it also affects
TCP timeouts (particularly in 9i RAC where there were no virtual IPs). After
some conversation via a TAR, Oracle did state that a lower value was acceptable,
as long as premature cluster evictions were not occurring.

So the question is, am I right in setting the heartbeat threshold to match the
misscount parameter? The value I am thinking of is about 30 seconds.

Thanks,
Gavin. 


-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://oss.oracle.com/pipermail/ocfs2-users/attachments/20051104/81d441a6/attachment.html