slow ftp and rcp between adjacent servers

[Tony Tibbenham]

We have 2 HP servers: 10.200.33.1 and 2
Both work OK. Both have different MAC addresses
Both have 10Mb cards disabled (via SAM) and 100Mb 1/2 duplex cards with the addresses above

.1 is HP-UX 10.20 and is working OK
.2 was working well under 10.20
.2 has just been upgraded to HP-UX 11
.2 seems fine except any attempts to copy files from .2 (rcp) is very slow and ftp'ing files from .2 to my pc is also slow.
Ping seems OK.
.1 will ftp to my pc at 1Mbyte/sec
.2 will only reach 16Kbyte/sec
.1 has 2 network cards, 100Mb EISA enabled
10Mb on-board disabled
.2 Has the same 100Mb EISA card and a 10Mb card.
10Mb on-board disabled
I'd suspect a routing issue except for the adjacent IP addresses and both servers being on the same switch.

Any ideas where else to look?

[A. Clay Stephenson]

These are rather classic symptoms of mis-matched duplex settings between the hosts and the switch ports. I suspect your config files in /etc/rc.config.d have been set to Auto-negotiate that us terribly unreliable. You should hard-set these cards to 100HD AND hard set the corresponding ethernet switch ports to 100HD.

[Santosh Nair_1]

Check that the 100BT cards are configured correctly, i.e. if the switch port is configured for 100 Full-Duplex, then configure the card the same way. Otherwise, the card will try to auto-negotiate the speed/duplex and will probably wind up with 100HD. Things will work, but they'll be extremely slow.

You can check the cards setting by using the lanadmin -x command, where ppa is the number listed under the Net-Interface/NamePPA column in the lanscan output. You can change the cards settings using lanadmin -X , e.g. lanadmin -X 100FD 0.

Hope this helps.

-Santosh

[V. V. Ravi Kumar_1]

hi,

set duplex at swith level and system level to same. better if it is a full duplex.ftp does perform better on full duplex.
regds

[Joseph C. Denman]

Hi Tony,

The above responses are correct. I have had the same trouble in the past, and it alway turned out to be duplexing not being the same on the switch and the hosts.

Hope this helps.

...jcd...

[James R. Ferguson]

Hi Tony:

I agree with Clay in that the first thing I'd look at is mismatched duplex settings between the network switch and the LAN card (as defined in the appropriate configuration file):

http://techsolutions.hp.com/fe/configs.html#config_files

Since you mention EISA cards, note that they support only half duplex (10 and 100mbps); and 'auto-negotiation' is not supported.

Regards!

...JRF...

[Tony Tibbenham]

All the replies above suggest switch / card duplex miss-matches unfortunately both card and switch agree on 100Mbit half duplex.
They both work well when re-install the 10.20 disks.
When running 11's netstat -p tcp we see:
500,000+ packets sent
500,000+ packets received
12,000 retransmit timeouts, 0 connections dropped by rexmit timeout.
On our working 11 box there are only 500 retransmit timeouts for similar packets sent and received.
Even telnet screen refreshes look slow: Is there some choke parameter that restricts bandwidth per connection?

[Jeff Arthur]

Hi,

Agree with all comments above.

The main source of these problems is the
settings for autonegotiation.

One end will probably trying to autoneg, and the other not.

This will lead to the poor performance you describe.

Here I don't have access to our switch configs,
(Different Team!!), but then I sun into these problems I jsut change things at the server end
and benchmark the result.

I change autoneg first, and if that doesn't fix it I try changing the speeds to get a match
(there aren't that many combinations)

You can use sam for ease or
lanadmin -X
from the command line

(usual settings are 10HD, 10FD, 100HD, 100FD
AUTO_ON )

hope this helps
Jeffa

[Chris Wilshaw]

I had a similar problem after upgrading to 11 (some telnet-based processes were slower than expected).

Try setting the -TCP_DELAY option in /etc/inetd.conf as below;

telnet stream tcp nowait root /usr/lbin/telnetd telnetd -TCP_DELAY

then restart inetd (inetd -c)

Chris

[Tony Tibbenham]

I've tried -TCP_DELAY, no improvement
Tried lanadmin -X 100HD 1, no improvement
Checked in swlist - both running the same software for the card (J2780BA)
Checked /etc/rc.config.d/hpeisabtconf on both servers: Some differences - working card was set HP_EISABT_SPEED[0]=AUTO_ON and HP_EISABT_DUPLEX[0]=-256, a depracated setting included.
Replaced hpeisabtconf with the working one & restarted server - now working OK.
Thank you for all your suggestions