I have been doing more investigation on this issue.  You are right about 
the ethernet drivers.  But it does not only happen when using ssh, it 
also happens at the console until a ping packet (or some other network 
activity) brings the race condition back to normal.

After disabling the two onboard e1000 network ports and installing a 
3c905 network card, the race condition no longer occurs.  So I believe 
that there is a problem with the e1000 kernel module causing a race 
condition.  I have only been able to verify this on the PowerEdge 750. 
If anybody else has an e1000 card and would like to test this it would 
be great, but I would wager this problem exists on all systems with the 
e1000 driver.

I guess for now, I will sit it out with a different network card 
replacing the onboard one until the upstream vendor releases a new 
kernel fix.

Greg

John Franks wrote:
> Does this happen only when using ssh?  If so, I would suspect issues
> with the ethernet drivers.  There have been problems with the drivers
> for some Broadcom devices.  You might check out redhat bugzilla.  
> 
> On Fri, 2005-05-13 at 09:08 -0500, Gregory Harris wrote:
> 
>>Hi Folks,
>>
>>SL4 works great on all of the Dell Optiplex GX280's I've tried it on. 
>>However, I'm having some rather mysterious behavior on a dell poweredge 
>>750.  Dell has said they will start certifying hardware and officially 
>>supporting the vendor's enterprise 4 operating system next month, so it 
>>may not be worth the effort to try to figure this out.  Anyhow, I 
>>thought I'd sent to the list and see if anybody else is having any 
>>issues like this.
>>
>>1.  Sometimes, processes just seem to hang.  I've noticed this over an 
>>SSH session, a lot of times when using the yum command.  I ssh in to the 
>>server and see the process still in the process table.
>>
>>2.  Doing a 'service iptables restart', the system stalls on the 
>>"Unloading iptables modules:".  If I go to a different window and ping 
>>the server, the iptables service restart resumes.
>>
>>These issues aren't sounding like userland issues to me, although they 
>>very well could be.  Here are the kernel modules the machine is using:
>>
>>Module                  Size  Used by
>>ipt_REJECT             10561  1
>>ipt_state               5825  14
>>ip_conntrack           45701  1 ipt_state
>>iptable_filter          6721  1
>>ip_tables              21441  3 ipt_REJECT,ipt_state,iptable_filter
>>md5                     8001  1
>>ipv6                  238945  16
>>parport_pc             27905  0
>>lp                     15405  0
>>parport                37641  2 parport_pc,lp
>>autofs4                22085  0
>>nfs                   200997  6
>>lockd                  65129  2 nfs
>>sunrpc                137637  9 nfs,lockd
>>dm_mod                 57157  0
>>button                 10449  0
>>battery                12869  0
>>ac                      8773  0
>>uhci_hcd               32473  0
>>ehci_hcd               31813  0
>>e1000                  82253  0
>>floppy                 58065  0
>>ext3                  118473  3
>>jbd                    59481  1 ext3
>>ata_piix               12357  0
>>libata                 44229  1 ata_piix
>>aacraid                44529  4
>>sd_mod                 20545  5
>>scsi_mod              116301  3 libata,aacraid,sd_mod
>>
>>Thanks,
>>
>>Greg
>>
>>

-- 

Gregory Harris
Network Specialist
Department of Mathematics
The University of Kansas
[log in to unmask]
(785) 864-7311