On 03/27/2012 08:19 AM, Baker D.J. wrote:
Hello,
We're running xcat Version 2.6.10 on our cluster, and I've been taking
a look at the network with tcpdump recently. We have in the region of
1000 compute nodes on the network, and they're all managed by xcat.
The volume of ARP related traffic on the network (I ran tcpdump on the
xcat master) seems very high to me, however I'm no expert in these
things and so the level of ARP activity may be reasonable. I ran
tcpdump for just 1 minute, and recorded in the region of 7,000 ARP
transactions. Is this reasonable for a cluster of 1000 compute nodes?
If anyone can please advise us in this respect then that would be
appreciated.
Best regards -- David.
i wouldn't be too surprised if 7000 arps/sec is normal in this case. one
thing you might want to look at changing is the values around how big
the arp table can be. 1000 addresses is more than the default settings
for the kernel to keep in the arp table for more than 30 seconds. so it
might be that everything is constantly re-arp'ing cause the arp table is
getting cleaned out. http://linux.die.net/man/7/arp has some info in the
/proc interfaces section. you should look at the gc_interval and
gc_thresh{1,2,3}. it might be a long shot but i once managed a system
that had ~2000 devices on one broadcast domain and he had to up the
sizes of gc_thresh{1,2,3} to get it so things played nicely.
mike
------------------------------------------------------------------------------
This SF email is sponsosred by:
Try Windows Azure free for 90 days Click Here
http://p.sf.net/sfu/sfd2d-msazure
_______________________________________________
xCAT-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/xcat-user
