[ofa-general] Re: [Query] ib add path record cache

[Sean Hefty]

> Please anybody just tell me about the idea of _distributed SA_ in
> short. Is it a pre-planed activity which is yet to be implemented with
> the OFED? or its just an extension of the sa cache pre-loading
> discussion?

I'm thinking of a distributed component that can perform a limited set 
of SA functionality.  The sa cache is close in that it can respond to 
path record queries via an API call.  If the sa cache could respond to 
actual PR query MADs, IMO it then becomes a very simple distributed SA.

This idea came from trying to decide on the best way to pre-load the 
cache.  By using a MAD interface, I think we get several advantages:

* The existing userspace MAD interfaces could be used, which avoids 
adding a userspace interface for just the cache.

* The existing code in the sa cache used to process PR query responses 
is re-used.  (I.e. I anticipate that the kernel changes needed to 
support pre-loading to be fairly small.)

* We have a framework that can be used to load the entire cache, add a 
specific set of PRs, and remove specific PRs.

* The cache becomes accessible from remote systems - both for loading 
the cache as well as for queries.

So, I think that using a MAD interface to preload the cache is a 
relatively simple change, but gives us additional flexibility.  And to 
be clear, I'm not suggesting that we implement additional functionality, 
just that we have the framework available.

- Sean