[Linux-cluster] Cluster heartbeat - minimum nic speed

[Michael Will]

Now is that a failure (not able to send heartbeats indicates not able to
run the original service as well),
or should that heartbeat stuff run with a higher priority than the
application?
 
Michael

________________________________

From: linux-cluster-bounces at redhat.com
[mailto:linux-cluster-bounces at redhat.com] On Behalf Of Coman Iliut
Sent: Tuesday, December 12, 2006 12:31 PM
To: linux clustering
Subject: Re: [Linux-cluster] Cluster heartbeat - minimum nic speed


We've had issues with heartbeats being lost but it was always because
one node was internally too loaded, and the process sending heartbeats
was not getting enough time to run and send those heartbeats.

You must figure out what is your network traffic from the apps you're
running and make sure you're under the H/W limit. Sending a heartbeat to
the other node should not be a problem if you're under the limit. 

Coman


On 11/16/06, David Elliott <david.elliott at shazamteam.com> wrote: 

	Hello
	
	We're running a 2 node rhas4 cluster with gfs and fibre attached
storage
	
	At the moment we have a pair of bonded 10/100 nics for heartbeat
and gfs
	locking communication (bonded in active/standby mode) 
	There are an additional pair of gigabit nics bonded for general
network
	traffic
	
	We've been having some performance issues with GFS which have
been
	investigating - mainly to do with slow file stat operations like
find, 
	and ls.
	
	a comparison of a data set (700,000 files in a single directory)
on gfs
	and then on ext3 is below
	
	# GFS
	[root at mrapp1 ~]# time ls /free0/partnerimport/data/soap-2/
>/dev/null
	
	real     17m10.035s
	user    0m8.220s
	sys     0m52.310s
	
	# EXT3
	[root at mrapp1 ~]# time ls /mr/sig/partnerimport/data/soap-2 >
/dev/null
	
	real    0m59.854s
	user    0m5.296s
	sys     0m0.662s
	
	
	Can anyone confirm whether using a 10/100 nic for heartbeat
would be 
	having an impact on performance , and whether it would be
advisable to
	ensure these are gigabit?
	
	Many thanks
	Dave
	
	--
	Linux-cluster mailing list
	Linux-cluster at redhat.com 
	https://www.redhat.com/mailman/listinfo/linux-cluster
	




-- 
Coman ILIUT

Mitel Networks
Ottawa, ON 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/linux-cluster/attachments/20061212/bf3d55c0/attachment.htm>