[ewg] questions about OFED 1.2 IPoIB bonding

Scott Weitzenkamp (sweitzen) sweitzen at cisco.com
Mon Apr 9 21:53:34 PDT 2007


Moni,
 
I've been testing IPoIB bonding on RHEL4 x86_64 and RHEL5 x86_64, and
have some questions.  This feature looks like it has a lot of potential,
thanks for adding it to OFED 1.2.
 
1) IPoIB bonding and IPoIB CM do seem to work together, but after
running ib-bond --bond-ip, I have to manually reconfigure IPoIB CM (both
mode and mtu) again, then increase the bond0 mtu.  It would be nice if
ib-bond took care of this for me.
 
2) I tried a port failover stress test overnight (flipping a port every
10 seconds) and got an HCA catastrophic error, how much failover stress
testing have you done while traffic is running?
 
3) Can I configure IPoIB bonding to start at boot time as part of the
standard /etc/sysconfig network scripts?
 
4) Do you have any plans to support load balancing in addition to
failover?
 
5) I've seen some erratic throughput with netperf using bond0 (no
failover happening), have you seen this?  For example:
 
Interim result: 2731.18 10^6bits/s over 1.00 seconds
Interim result: 2732.57 10^6bits/s over 1.00 seconds
Interim result: 2717.93 10^6bits/s over 1.01 seconds
Interim result: 1609.63 10^6bits/s over 1.69 seconds
Interim result:  487.73 10^6bits/s over 3.30 seconds
Interim result:  394.51 10^6bits/s over 1.24 seconds
Interim result:  380.00 10^6bits/s over 1.04 seconds
Interim result:  621.87 10^6bits/s over 1.08 seconds
Interim result:  372.02 10^6bits/s over 1.67 seconds
Interim result:  388.15 10^6bits/s over 1.00 seconds
Interim result:  419.81 10^6bits/s over 1.00 seconds
Interim result:  369.75 10^6bits/s over 1.14 seconds
Interim result:  858.04 10^6bits/s over 1.00 seconds
Interim result: 2285.79 10^6bits/s over 1.15 seconds
Interim result:  501.81 10^6bits/s over 4.56 seconds
Interim result:  465.64 10^6bits/s over 1.08 seconds
Interim result:  499.98 10^6bits/s over 1.00 seconds
Interim result:  478.41 10^6bits/s over 1.05 seconds
Interim result:  482.03 10^6bits/s over 1.00 seconds
Interim result:  494.85 10^6bits/s over 1.01 seconds

 
Scott Weitzenkamp
SQA and Release Manager
Server Virtualization Business Unit
Cisco Systems
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ewg/attachments/20070409/840b4713/attachment.html>


More information about the ewg mailing list