So far, I was waiting to confirm, but I think the ethernet configuration is
the issue, as suggested
by 99% of people responding. Thanks to all of you - replies follow the
original messages:
-----Original Message-----
From: Riggs, Joan [mailto:JRiggs_at_kls.usaka.smdc.army.mil]
Sent: Friday, March 16, 2001 8:10 AM
To: 'tru64-unix-managers_at_ornl.gov'
Subject: large FTP fails
So far, Compaq has not been able to help me - perhaps someone here has had
this strange problem before:
We have recently been unable to FTP large files from one of our ES40s to
another. Small files transfer just fine,
but large ones (say, >1 GB) get so far and just hang.
We were able to prove it was hardware related by swapping hardware from one
system to another and the
problem moved. Compaq thought it was the PCI backplane so we replaced it
but the problem still exists.
We drop virtually all incoming packets - outgoing is fine.
Any ideas? I am out of them and am about to call Compaq hardware support
back and argue with them again.
Thanks in advance,
Joan Riggs
Replies:
George Gallen
what hardware was swapped that caused the problem to move? Were you able
to isolate down to one card. When the hang happens, Does the hub/switch on
the sending side show any collisions.
If the problem moved when swapped, and if you could isolate it down to one
piece,
have you tried replacing that one piece?
me:
We replaced the PCI daughterboard and it did not fix the problem. I am not
really a hardware person,
so am having to trust Compaq support on this one. We're in the middle of
the Pacific so they can't
dispatch an field engineer to fix it, either!
George Gallen again
Will the opposite work? You said when you moved equip the problem
followed.....Did it
happen before you moved the equipment, that is to the one you moved it to.
If it didn't
and you move it to the non working one does the non working one now work?
Basically, the question is did you receive a dead or non-working board as a
replacement?
me: NO
O'Brien, Pat
seen similar issue when duplen of network card was wrong or duplex value of
switch( in ourcase) was set incorrectly.
Tom Linden
How small is small? Swapping the hardware doesn't prove it was the
hardware. It actually sounds like a software problem.
Are there some size limitations related to you ftp setup?
me: NO
Gary Phipps
What hardware did you swap ???
We have experienced similar problems when the nic is not set to the correct
speed (ie half duplex).
Jean-Louis Oneto
I'm not sure it can apply, but I often noticed that network adapters are
unable to manage packet fragmentation ; a way to test that is to use the
ping command with increasing packet size. I'm unable to explain this
behavior at hardware level, but I have a lot of such network adapters which,
most of them after a storm, work perfectly with small packets (ie ~< 1500)
but hang or drop larger ones...
Klas <mailto:Klas.Erlandsson_at_europolitan.se> Erlandsson
Are you using a GigabitEthernet card?
me: NO
Weirick, Robert
You don't say what hardware you swapped? Did you move the NIC?
Also, have you looked at the interface (netstat -i tu0) to see if there were
an inordinate amount of dropped
/error packets?
me: yes, there are a lot on dropped incoming packets only. outgoing is
fine.
John P Speno
Sounds like a duplex mismatch on your fast ethernet network.
Don't use autonegotiation on either end. Set the switch and host
explicity to 100MB full duplex and re-try your ftp.
me: still waiting on our network guys but turned off auto-negotiate on my
servers with lan_config.
Allan E Johannesen
Check to be sure that half/full duplex settings agree with the network
equipment. Auto-negotiation sometimes fails, if that's being used. It might
be best to specify.
Robert Fridman
Try installing another network card and see if it solves your problem.
Richard Jackson
I don't think it would cause a total hang but verify your switch and NIC are
both in half or full duplex.
Richard Westlake
Have you checked the fuill/half duplex settings on the ES40 and switch
ethernet ports.
Anything interesting in the syslog messages .
Mandell Degerness
Is it your MTU on the card?
Try setting both the sending and receiving ipmtu to some safe number (i.e.
1400) and see if that helps. I had to do that for some problems I had a
couple years ago.
me: how would I do that?
Don Wilkerson
I ran into this same thing on an HP/UX system, but did'nt look at hardware
or anything like that. I needed to get the file moved so I tried 'rcp'
instead and that worked.
I'm wondering if rcp would do the same for you?
me: I need to be able to ftp also.
Phil Baldwin
Just a thought. it couldn't be anything to do with autosensing 10Mb/100Mbs
network cards? I have known where the autosensing feature can cause
interesting problems.
Knut Hellebų
Is this just one system failing to ftp >1GB files or is it several ?
me: all three ES40s
Joe Fletcher
This isn't something as basic as a duplex mismatch on the NIC is it?
When you say you swapped hardware which bits did you move?
me: it was that basic, thanks.
William H. Magill
It's a hardware problem alright, but on the network.
You are trying to run at 100 meg/full-duplex etherent speeds correct?
You need to hard set the duplex mode on the switch and the CPU (at the SRM)
for full-duplex or you will get a duplex mis-match, and exactly the results
you are seeing.
The primary symptom of the duplex mismatch is exactly yours. Small files
transfer fine, large files go to ...east-japip... None of the error counters
in the switch or the host increment on this problem, you just drop packets.
The auto-negotiate feature doesn't work for squat (ie doesn't
inter-operate) with almost every vendor out there. You can auto speed
detect with no problems, but the duplex negotiation doesn't work.
And it is NOT a Compaq problem, it is an industry problem. We have several
hundred (maybe thousands) of switches now in our network and do not let any
of them run in auto mode, it saves us LOTS of trouble tickets.
Bryan Williams
It might be as simple as the NIC card. I have seen this happen in other
large network volume applications.
Do a netstat -in
See if you have any Ierrs, Oerrs, or Coll (Input errors, Output errors, or
Collisions).
If you have any errors, there are a couple of possibilities. The most
common one I've seen is a NIC and the switch/hub being set
inconsistantly, i.e. one is half duplex and one is full duplex.
If you hub/switch thinks you should be full duplex and you're not, then
in /etc/rc.config, look for the line that says IFCONFIG_x . It will have
your ip address and the netmask. Put "speed 200" after the netmask value
and reboot.
It could also be wiring, the hub/switch, or an old/bad NIC card.
You want to check this on both ends of your ftp.
Phil Baldwin
Its autosensing.
Problems can occur where you have a network card to autosense it's LAN
network connection rate. Sometimes with some LAN switches / hubs, they have
difficulty deciding what speed to run at, leading to unpredictable problems.
If you suspect this is the cause, set the speed on the network card to
10Mb/s half duplex (this can be done in System Configuration).
Received on Wed Mar 21 2001 - 03:00:47 NZST