[ofa-general] Re: [PATCH 1/4] opensm/osm_ucast_cache.{c, h}: ucast routing cache implementation

[Hal Rosenstock]

On Tue, 2008-05-06 at 00:58 +0300, Yevgeny Kliteynik wrote:
> Hal Rosenstock wrote:
> > Hi Yevgeny,
> > 
> > On Sun, 2008-05-04 at 12:59 +0300, Yevgeny Kliteynik wrote:
> > 
> > I haven't yet had a chance to review this in detail but think that
> > router ports need to be accomodated in the subnet (I think this is a
> > firm requirement as router ports on the subnet are already supported)
> > and also think that nothing should be introduced precluding the running
> > of OpenSM on a router port. From the latter standpoint, it looks much
> > like a CA port.
> 
> This is exactly how I implemented it - any non-switch port is
> treated as CA, which is just a target LID.
> 
> Well, I mean I intended to implement it that way - I reviewed it again,
> and it appears that the cache is fine with routing to routers and running
> from switches, 

then it's just the variable names which indicate ca if routers are
grouped with cas.

> but there will be a problem when SM runs on a router node -
> cache will complain and fall back to usual routing.
> That can be easily fixed.

Right; the one thing I saw was in _ucast_cache_get_starting_osm_sw where
routers were not supported. I think a one line change is all that's
needed there. Not sure if there are other places.

> However, I'm not sure how the OpenSM will behave in general when running
> from switch or router - I've never tried it. Has anyone try it?

I'm not sure either but would be interested to hear. I think there are
some using it on switch port 0 and also think others have tried it on
routers. In terms of switches, it used to work and there is some support
in the vendor directory for this.

-- Hal

> -- Yevgeny
> 
> > -- Hal
> > 
> > 
>