[realtek] NFS problem (using realtek)

Lynn Winebarger lynn@freespeech.org
Fri, 20 Oct 2000 20:49:43 -0600 (MDT)


   I've got a little 4 machine network, with one machine an nfs server and
another running an active webserver (on a separate interface to the
outside world).  The other 2 machines aren't currently being used much
(but will be).  All the NICs hooked up to the 4 machine network are DLink
530TX's with rtl8139 chips.  
   This morning I was copying a few hundred megabytes worth of files over
NFS from machine #3, and the webserver machine's load went through the
roof (because all the httpd processes were in an uninterruptable sleep,
waiting on NFS).  I tried pinging the nfs server, and got back responses
in pairs, where one would be about 0.3 ms, and the other 1000 ms.  I shut
down the copying process, waited a couple of minutes, and pinging the nfs
server got the same results.  I waited a few minutes more, tested again,
same results, rebooted the server, worked fine. (machine #3 didn't
experience any of these problems, though it also uses the rtl8139 driver).
   Anyway, I see similar problems are reported for the rtl8139 driver in
the archives.  I'm using Redhat 6.2, with kernel 2.2.16-3 on the
webserver.  Looks like the rtl8139 driver is 1.07.  Would upgrading to
1.10 fix this problem?  Is there a fix in the near future?
   I'm going to be building a small cluster of webservers using this nfs
server, and buying a bunch of NICs, so I'd like to know whether avoiding
realtek 8139 based cards is necessary (looking at the tulip list, looks
like they're not suffering from the same problem).

Thanks,
Lynn