[K12OSN] nfs troubles

["Terrell Prudé Jr."]

Craig White wrote:
> On Fri, 2007-11-16 at 11:18 -0600, Jon Harder wrote:
>   
>> I am having problems booting iMac clients that I believe
>> is related to overwhelming them with data.
>>
>> This is what I am seeing as they boot up:
>>
>> ...
>> Mounting root filesystem: /opt/ltsp/ppc from: 192.168.0.254
>> Doing the pivot_root
>> nfs:server 192.168.0.254 not responding, still trying
>> nfs:server 192.168.0.254 OK
>> nfs:server 192.168.0.254 not responding, still trying
>> nfs:server 192.168.0.254 OK
>> nfs:server 192.168.0.254 not responding, still trying
>>
>> and these last two lines repeat indefinitely.
>>
>> Some searching reveals that this might be related to the old
>> computers being overwhelmed with too much UDP nfs traffic from
>> the server.
>>
>> Originally I tested this setup with a few iMac clients and it
>> worked fine. Now that I want to roll it out, I purchased a new
>> HP ProCurve 1800-24G switch which is apparently dumping the data
>> from the server's Gb connection a lot faster than the switch in
>> my original test setup was able to do.
>>
>> I hooked six of the iMacs to an old 8 port switch and they all boot
>> fine. Once they were up and running, I moved the connections over to
>> the new fast switch and they work better, but soon the oldest iMacs
>> freeze up.
>>
>> One suggestion I found was to force the clients to connect to NFS
>> using TCP instead of UDP. I tried playing with 
>>
>>   /opt/ltsp/ppc/etc/rc.sysinit
>>
>> a little bit to force TCP, but with no apparent success. I use
>> tcpdump to monitor the network. The server is pushing out a lot
>> of UDP packets!
>>
>> Can someone suggest what I might do to ensure the clients are using
>> TCP? Or is there a way to throttle the server so that it only sends
>> so much traffic to a particular client? Any help would be appreciated.
>>     
> ----
> Jon,
>
> I had terrible problems at first and found that the spanning tree
> protocol on my Dell switches had to be disabled...I found no other way
> around this (reliable way anyway).  The way I discovered/proved the
> problem was to make sure that the last switch between client and server
> didn't have spanning tree protocol...i.e...
>
> Server -> cheap switch -> expensive switch with spanning tree -> 
>                                                        cheap switch ->
>                                                               mac client
>
> Or just eliminate the expensive switch entirely from the equation and it
> worked fine. 
>
> Then the issue is simply to disable the spanning tree protocol on your
> switch if that's the problem
>
> Craig

Yep, older Macs have always had challenges with the switch running the
full spanning tree algorithm.  It is known to play havoc with DHCP on
Macs in particular.

Some switches might not let you actually disable the spanning tree
protocol (STP).  Cisco IOS-based switches, like 3550's, number among
these.  However, you can put it into what's called "fast learning"
mode.  Basically, this means two to four seconds instead of nearly 30
for the port to come up once you fire up the iMac client.  Works like a
champ.  My iMac Bondi was much happier afterwards.

--TP
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/k12osn/attachments/20071116/d7c34a24/attachment.htm>