[ofa-general] Some problem about the LID selection.

[Todd Rimmer]

The intent of the specification was for the SM to provide such suggestions, hence fabric policies and config could be easily controlled and path selection can be coordinated with routing and topology knowledge (which the SM has).

Generally Path Record queries should be made using SGID and DGID.  In which case the SM can return multiple matching paths.  It is up to the SM to return an appropriate subset of paths.  The SM can also return multiple paths  in "preferred order of use" (eg. Primary, then secondary path for failover, etc).

The NumbPath parameter in the PathRecord query can specify the maximum number of such paths of interest.  Typical  use cases are:

-          Ask for 1 path (most applications)

-          Ask for 2 paths and perform a failover algorithm or use APM (I'm aware of quite a few applications which use failover)

-          Ask for many or all paths and perform a load balancing algorithm across them (I'm aware of a couple applications which do this)

Todd Rimmer
Chief Architect
QLogic Network Systems Group
Voice: 610-233-4852     Fax: 610-233-4777
Todd.Rimmer at QLogic.com<mailto:Todd.Rimmer at QLogic.com>  www.QLogic.com<http://www.QLogic.com>

From: general-bounces at lists.openfabrics.org [mailto:general-bounces at lists.openfabrics.org] On Behalf Of Jordan
Sent: Tuesday, July 14, 2009 7:26 AM
To: general at lists.openfabrics.org
Subject: [ofa-general] Some problem about the LID selection.

When LMC > 0 , there are 2^LMC LIDs assigned to a port.  When choosing different LID, there maybe different paths to get to this port.
In other words, different LIDs represent different paths. But which path should I choose to  get to the destination? It seems that OpenSM
does not provide such algorithm .

Does OFED provide such algorithm to manage  which path should to choose ?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20090714/4e306114/attachment.html>