Bill,
On Mon, 2005-01-10 at 15:15, Bill Wichser wrote:
In answer to my own question, the problem can be traced back to a software issue with the HP switch. By turning off LACP on all the ports, the timings work so that the installation continues.
Chalk one up for HP then. Switches from SMC and Nortel Networks, no matter what you turn off or on (fast forwarding, edge ports, spanning trees, etc.) are still afflicted by this dual-DHCP-request problem. I've seen it since the RH9 days. It always took an anaconda hack to fix and rebuild the first stage loader. RH has supposedly taken care of it in 3.4, which I'm just now testing. We'll see if it works.
Bill
Bill Wichser wrote:
There appears to be an issue with installation over an HP switch. Upon bootup, the network driver resets twice which somehow never allows the client to ask and therefore receive it's MAC address.
This follows the RH bug : https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=110036
Installation via a directly attached cable to the head node works fine.
I have emailed the Beowulf list in hopes that the problem was in some configuration of the switch but have recently come across this bug report and it appears that the symptoms are the same.
Are there any plans to roll these patches into CentOS 3.3 (x86_64) in the near future?
Thanks, Bill
CentOS mailing list CentOS@caosity.org http://lists.caosity.org/mailman/listinfo/centos
CentOS mailing list CentOS@caosity.org http://lists.caosity.org/mailman/listinfo/centos
Will Dinkel Chief Technology Officer Team HPC, Inc. 785-542-2135 x304 wdinkel@teamhpc.com http://www.teamhpc.com