Description of Problem: The upgrade to the latest errata kernel 2.2.19-7.0.1 caused a switch to nfs3, however, this does not seem to work stable! If the load gets to high the sever will go down approx. once per hour: Either the eth intercae will be gone (can be brought up again from the console), or the kernel will panic. I disabled nfs v3 on the server and now the server runs stable again since almost a week. How Reproducible: Since this is our main server, I cannot do any furhter testing. Additional Information: Hardware: AMD Athlon(tm) Processor 770MHz, Realtek RTL-8139 (rev 10) ethernet card
It's more related to troubles with RTL8139 driver, which have EXPERIMENTAL status in kernel and very buggy on high-load ( your may go to www.scyld.com - homepage of Donald Becker (original version of this driver). Also SCYLD host maillist realtek-bugs - your found many bugreports, like your :) And bugs with rtl8139 often with NFS (UDP high load). Interface stop respond or system reboot :) Alternative - use Donald Becker latest rtl8139.c or alternative driver 8139too (originally developed for 2.4, but then backported to 2.2 and included with >2.2.18). But may choise - replace 8139 with more stable netcard(stable driver) ( Intel EE100 for example). Sorry, my English very poor :)
It might indeed be 8139 related. Could you try to use the 8139too module instead? (just change the modulename in /etc/modules.conf)
Thanks for your help. I wasn't aware of the other module. In fact, I changed it yesterday evening and there were no problems since. (However, I have not turned nfs v3 on again). I'll keep you informed.
I assume that everything is ok, and will close this bug as "Currentrelease". If you have problems again, please reopen this bug.
Yep, the 8139too module seems to wor better, but it is MUCH slower! Anyway, I'll move to 2.4.