[openib-general] Some ib_mad.h Redirection Comments

Yaron Haviv yaronh at voltaire.com
Sat Aug 7 12:29:18 PDT 2004


On Saturday, August 07, 2004 7:08 PM, Sean Hefty wrote:
> On Sat, 7 Aug 2004 01:09:58 +0300
> "Yaron Haviv" <yaronh at voltaire.com> wrote:
> 
> If we're in agreement on the server side, then I think we're more than
half-way
> there.   

To me it looks like there is agreement :)
 
>> I don't see what is the value in having every potential GSI client
>> implement a common functionality of identifying it's a Redirect, and
>> resending the request to the new location, why not just have the GSI
>> layer do that common functionality for all the consumers ?
>> I think it is also trivial for the GSI to implement such
>> functionality 
> 
> My general thought is that if something is trivial to implement, just
> push it up to the clients.  However, I'm not sure that redirection is
> that simple.  With a 1000 node fabric, 3 services per node, and
> everyone redirecting, the redirecting table will be substantial.  I
> need to think about this more and continue discussing it, but I'm
> becoming convinced that there may be enough work there to justify
> moving some of this into the access layer.      
> 
>> If your main concern is CM (that has a single client per node,
>> multiple remote servers, and some different behavior) we can treat it
>> as an exception, I'm more focused at things such as distributed SA in
>> such cases we can cache/lookup based on a class first
> 
> I pick the CM because it's the simplest type of client.  It doesn't
> use RMPP or request/response transfers.  I think starting with the CM
> will ensure that we get the right layering of services.  

For CM the use cases are different than SA, PM, .. Since CM Redirect can
be used for per session load balancing, etc', in such a case Redirect is
better done at the CM rather than the Access 
For SA .. There is usually only one target (SA Server) accessed by each
node, so we just need to cache one entry for the remote SA
One implementation I'm looking at is a distributed SA, were each node
access a different SA, also there can be more than one SA clients per
node Vs. usually one CM per node   
So that's why I think Redirect in CM & SA are not the same and may
require different approach 

Your thoughts ?

Yaron 



More information about the general mailing list