Adam,
Don't see your problem but we use a mix of z/OS and Linux (SuSE 2.4.17) so
can't exactly match your config.

A few basic questions:

The Linux guest that "owns" the OSA - does this mean the OSA is not shared
at all, and/or have you defined it as PRI router?
Are the Hipersockets addresses added to /proc/qeth_ipa_takeover ?
Are you using dynamic routing at all (Zebra / gated)?
What about MTU sizes and/or are you using Path MTU discovery on all stacks
involved?

Ciao
Mark


-----Original Message-----
From: Linux on 390 Port [mailto:[EMAIL PROTECTED]]On Behalf Of
Adam Thornton
Sent: 01 July 2002 16:25
To: [EMAIL PROTECTED]
Subject: Strange TCP (?) behavior with OSA-GigE/HiperSockets


I've got a situation where I've got a Linux guest owning an OSA-GigE,
with (virtual) HiperSockets networks routed behind him.

Ping times to the HiperSockets guests are effectively instantaneous.

If I ssh to the guest with the Gig-E and then SSH to the guests behind
him, *that* interaction is instantaneous.

But if I SSH directly to the guests on the HiperSockets, it works but
the connection is unresponsive and jerky; it feels like I'm typing over
a transatlantic line.  Everything gets there eventually but it sometimes
takes a few tenths of a second.

I'm guessing that this is something to do with TCP buffering packets as
they pass from one interface to another, since the problem doesn't exist
with ICMP packets.  Further I suspect that--since I'm running the SuSE
2.4.7-timer kernel--it may have something to do with TCP not flushing
its buffers on each timer tick, since there are no timer ticks.  This,
however, is a guess which I have not yet attempted to construct a test
case for.  Has anyone else seen this behavior?

Adam

Reply via email to