Solaris 10 8/07, Broadcom PC NICs, poisoned arp tables and RPC 826
See the opensolaris.org discussion here.
If you have Solaris 10 Update 4 machines experiencing intermittent hangs or inability to communicate, only to clear up and re-occur, time to rule this in or out. It's pretty easy to identify if you are running into this:
for example:
capture "netstat -p" to a file when communication is normal and then again when the communication is hung, (but not yet recovered). A diff will show you this:
# diff /tmp/netstat_p4_working /tmp/netstat_p4_notworking< e1000g0 123.45.67.100 255.255.255.255 00:14:4f:2d:6c:86
> e1000g0 123.45.67.100 255.255.255.255 00:15:17:3d:bc:2c
If you find incorrect MACs, they will most likely start with an Intel or Dell vendor code. Those intrusive MACs will belong to a PC using a Broadcom NIC in "teamed" mode. (The opensolaris discussion link above has a snapshot of this in Ethereal.)
Solaris will conform to RFC 3927 and will "defend" it addresses. When that happens, the problem will resolve. Until the next time the teamed NIC interfaces send out bogus info. As an immediate workaround, unconfigure "nic teaming" on the affected Windows interfaces. Contact your PC vendor for an updated driver.

Hi ML Starkey,
Thanks for your update. I think we are facing the same problem in our Environment. So, we hope to know if there any Solaris 10 patch available for this situation.
Thanks & Regards
Paul
Posted by Paul Liong on March 11, 2008 at 03:55 AM EDT #
No patch from Sun, as the flaw is not with Solaris. You have to update the PC drivers to solve the problem (or unconfigure the nic teaming).
Posted by ML Starkey on March 11, 2008 at 06:04 AM EDT #
Hi ML Starkey,
Thanks a lot.
Regards
Paul
Posted by Paul Liong on March 11, 2008 at 09:07 PM EDT #