[openib-general] SA MADs and Cisco SM

[Roland Dreier]

 > I am developing an API that uses SA MADs to create and get 
 > ServiceRecords.  I am using the OFED1.1 mad and umad libraries.  
 > Everything works flawlessly with OpenSM.  However if I run in a fabric 
 > that is using the SM embedded in a Cisco/Topspin switch, queries 
 > (SubnAdmGetTable) for ServiceRecords fail with a status 110 (ETIMEDOUT).
 > Since I don't have direct access to the switch and logs I can't tell 
 > what is going on from the switch, I am working on getting those logs 
 > asap.  However I was wondering if there were any known issues with 
 > interoperability, or functionality with OpenIB and the Cisco SM?

I don't know of any issues with the Cisco SM, and I do most of my
development using the Cisco SM running on Cisco switches.

However, since you are seeing a problem, it would probably make sense
to work with Cisco support to figure out if there is an issue with the
embedded SM on Cisco switches.  In this case, since the error is
ETIMEDOUT, it might make sense to try your query with a longer
timeout; it could just be that returning a big table is taking longer
than the timeout you set.  Perhaps opensm works because it's running
on a fast server CPU, while the switch SM is running out of gas on the
embedded CPU in the switch.

 - R.