Experimental Physics and
Industrial Control System

Matthias Clausen <[email protected]> · Sat, 16 Mar 2002 21:58:51 -0800

Hi Steve,

thank you for analyzing the error message Slava received on his PPC.

On Friday we found a problem in our installation which might be worth mentioning:

The current setup is:

EPICSF2 is the alias name of our SUN cluster (krynfs-a and krynfs-b).
The cluster provides automatic fail over. IOCs are booting from the cluster
through the alias name.

Since we had serious performance problem (also seen as long boots for IOC's) we
called a technician and found the problem on Friday:
The SUN cluster is connected to a Cisco switch. This switch is connected (100BT)
to a 3-COM switch which connects (100BT) to a 3-COM corebuilder and this way to
our FDDI backbone. The connection (100BT) 3-COM 3-COM has to be kept because this
is the only one proven to work properly with 100BT.
The problem:
The 3COM is of the year 1996 - Cisco from 2001 ;-()
Auto negotiation of the full/half duplex functionality between the two is seen
differently:
3-COM Full duplex / Cisco Half duplex -> must be a problem! In this case the
bandwidth dropped to 10% with lots of collisions.
The solution: manually setting the Cisco to Full duplex - works.

So be aware if you connect old equipment with new stuff - they might not
understand each other!

Matthias

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System