[openib-general] [PATCH 0/4] opensm: Loading unicast routes from the file

[Sasha Khapyorsky]

Hi Eitan,

On 09:36 Sun 11 Jun     , Eitan Zahavi wrote:
> Hi Sasha,
> 
> General comments:
> 1. I hope the change in osm.fdbs is not going to break the parser in
> ibdm:Fabric.cpp -

The file format was not changed, I don't expect brokenness.

> was it really necessary change?

Yes, in order to create unified osm.fdbs with any routing engine.

> or just nice to have ?

This is the nice side effect.

> 2. The modular routing is a great idea. From my first glance it seems
> that it assumes calculation of min-hop-tables is common to all routing
> engines.

Yes and no. Currently the min-hop-tables are used with multicast, so it is
common code. But I expect this will be different in the future (for
instance extend this loader to handle multicast tables too).

> I think it should be a callback provided by the engine too.

Yes, when it will be useful.

> Please note that the Min-Hop engine takes most of the routing time so in
> the future if we could avoid that stage it would be even better. 

Agree.

Thanks for the comments.
Sasha

> [EZ] We should start thinking about testing of this new feature too.
> 
> Further comment on the patches themselves.
> 
> > There are couple of unicast routing related patches for OpenSM.
> > 
> > Basically it implements routing module which provides possibility to
> load
> > switch forwarding tables from pre-created dump file. Currently unicast
> > tables loading is only supported, multicast may be added in a future.
> > 
> > Short patch descriptions (more details may be found in emails with
> > patches):
> > 
> > 1. Ucast dump file simplification.
> > 2. Modular routing - preliminary implements generic model to plug new
> > routing engine to OpenSM.
> > 3. New simple unicast routing engine which allows to load LFTs from
> > pre-created dump file.
> > 4. Example of ucast dump generation script.
> > 
> > Please comment and test. Thanks.
> > 
> > Sasha