[ewg] IB HA mode bonding over SDP

[Richard Croucher]

The bonding driver only works for IPoIB.   If  you are have writing you own
VERBS program then you can use Automatic Path Migration (APM) but this does
require not insignificant coding.   

 

Currently SDP, RDS, SDR, iSER don't include any failover.   Some, such as
SDP and iSER at least, use IPoIB as a namespace; using it to identify the
end-points with the familiar IPv4 naming tools.  The currently active link
from the IPoIB bonded pair will therefore be used for this naming unless you
specifically use the non-bonded IP addresses to identify the interface.

 

Failover is not such as issue for the storage protocols, since most
platforms already have solutions for channel failover for block storage
devices.  On Solaris, I used MPxIO successfully over SRP.  

 

Whilst SDP does not currently include APM support it will automatically fall
back to IPoIB in the event of a problem.

 

APM is relatively new to OFED.   I know of a number of VERBS programs which
have specifically coded it in.  

So the bottom line is that OFED provides the capability, but it is up to the
ULP protocol to decide if to use it and to then add the necessary code. 

 

Richard

 

 

From: ewg-bounces at lists.openfabrics.org
[mailto:ewg-bounces at lists.openfabrics.org] On Behalf Of bhavinkumar thumar
Sent: 18 March 2011 09:11
To: ewg at lists.openfabrics.org
Subject: [ewg] IB HA mode bonding over SDP

 

Hi All,

I am a newbie in infiniband domain. I am trying to use IB over SDP instead
of IPOIB. My problem is I have got 2 Mallenox IB cards on a machine so I
created a bonding using ib-bonding driver. I have installed OFED 1.5.2. But
with this bonding, connection is getting disconnected whenever link failure
occurs.

Is there any solution to this problem? I look around the web for the same
and I found that the HW failure would cause the IB RC session of a native IB
ULPs (SDP, RDS, iSER, Lustre, rNFS) to break, but I haven't found any
solution for this. Do I need to change anything in OFED?

Please help.

Thanks & Regards,
Bhavin

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ewg/attachments/20110321/642dfdbb/attachment.html>