Re: Intermittent network failure -- MIDRANGE-L

Jerry,

Interesting... Every other post I have from the archive says that both
ends should be hard configured because the negotiations don't go smoothly.

I understand the power savings warning, but in this case where it can
and does hit several PCs at the same time if they are actually working
at the wrong point in time I doubt if it is related.

Thanks.

Roger

On 4/3/2007 9:57 AM, Jerry Draper wrote:

We had a similar problem except that it wasn't CA dropping it was the i5
itself. We were configured at 100MB on the 400 and *auto on the switch.

IBM had us apply a series of microcode ptf's to the ethernet adapter.
They recommended to put the i5 at *auto and the switch at *auto.

No problems since.

Jerry

PS: Beware that power saving options can cause emulation drops so I
leave "turn off disks" and "standby" at never.

J

Roger Vicker, CCP wrote:

Before I spend $2000 for new switches (probably Netgear's Prosafe Smart
Switch series) has anyone got any diagnostic hints on this problem.

Model 800 at V5R3 with a Gigabit card (configured as 1G *FULL) to a SMC
6624M stacked switch's Gigabit port (configured as Auto-1000, the only
Gigabit option). XP Pro PCs running Client Access Express (CAE)
connected to the stack's 10/100 ports. New Cat5e patch cable tested with
a medium-range network cable tester and installed for the Gigabit
connection.

Every day at least once, any CAE session that is actively talking at the
wrong instant (not just sitting on a screen but sending/receiving) will
fail. The system doesn't log anything to Qsysopr and in fact thinks the
session is still valid. The CAE session seems to freeze for about 5-10
seconds then goes black and thinks it is disconnected. If you tell the
CAE session to connect it gives an error that the device name is in use.
The only solution is to wait until the TCP timeout expires (5 minutes)
or to use another session to DSCJOB the interactive job attached to the
session.

The switch's event log shows nothing. The port counters show zero for
any of the bad things like drops and errors.

It can hit multiple PCs at the same time if they are at an active
send/receive function.

The time of day that it happens is not really predictable. Although it
seems to cluster. Might hit for 5 minutes in the morning and not again
till quiting time. It seems to only effect emulation (only protocol that
is heavily used, no streaming audio/video...) but is not limited to IBM
CAE as I am using Rumba and I get hit too. The one difference with Rumba
(which does all communications through a single Engine) is that other
sessions on the PC will usually go in to *DSC status automatically. Just
the unlucky session that was talking at the wrong moment is left fully
active on the system. The Rumba Engine reports a communications error
and shuts down. Once the falsely active session is cleared all the Rumba
sessions and Engine connect like normal.

Thanks.

Roger Vicker, CCP