[ofa-general] Re: bug 400: ipoib error messages

Michael S. Tsirkin mst at dev.mellanox.co.il
Wed Mar 14 00:49:14 PDT 2007


Quoting Scott Weitzenkamp (sweitzen) <sweitzen at cisco.com>:
> The IPoIB failover is very slow sometimes, shown below is netperf -D
> output.  IPoIB failover should ideally only take a second or two.  I'll
> be filing a bug for that.
> 
> Interim result: 4355.09 10^6bits/s over 1.00 seconds
> Interim result: 4371.07 10^6bits/s over 1.00 seconds
> Interim result: 4370.95 10^6bits/s over 1.00 seconds
> Interim result:  162.41 10^6bits/s over 26.91 seconds

Scott, what is "Interim result"?
You are using SM on switch, are you not?
Are you sure the delays are not due to SM?

For failover to take place, the following needs to happen:
1. link down and notification - triggered by SM MAD
2. interface down
3. interface up (includes registration with SA)

2 out of 3 steps involve SM/SA

Since ipoib ha is just a perl script, it should be easy for you
to add logging there so you can figure out where's the delay

-- 
MST



More information about the general mailing list