[PHANI] inline below
________________________________
From: Chad Tindel [mailto:[EMAIL PROTECTED]
Sent: Friday, December 21, 2007 9:14 PM
To: Marella P-G19460
Cc: [email protected]
Subject: Re: [Users] Instable cluster with CPU load
IMHO, the later case may be considered ok depending upon a
node's design. A CPU-bound infinite loop can significantly starve other
user-level processes in the node. Such a starvation of other user-level
processes (leading to intra-node healthcheck timeouts, say) can actually
be considered a fault.
There's absolutely no reason why a non-realtime process should be able
to starve the heartbeating mechanism of an HA cluster simply by going
CPU bound.
[PHANI] Firstly, yes, I must confess that the my statement above is
incorrect; Being CPU bound is harmless by itself. It has to be entirely
something else which makes it harmful such as it being realtime or it
having higher nice priority (such as through 'nice -n -10').
And, I meant intra-node healthcheck timeouts (because they have to
compete for the same OS's attention) and not clusterwide heart-beat.
I'll see if I can get the report from the HP team sent to the users
list.
[PHANI] Ok, lets check that out. None of my earlier excuses seem to fly
;-)
Chad
_______________________________________________
Users mailing list
[email protected]
http://list.opensaf.org/maillist/listinfo/users
