[openib-general] Question

[Hal Rosenstock]

Hi Ron,

On Mon, 2005-02-28 at 17:38, Ronald G. Minnich wrote:
> On Tue, 1 Mar 2005, Yaron Haviv wrote:
> 
> > Ron, I believe netdiscover uses direct route MADs
> > So it can work also when the fabric is not fully initialized 
> 
> ok, that makes sense. 

but it's incorrect.

> So this brings up another question. ibnetdiscover is plenty fast, and 
> opensm is plenty slow.

OpenSM is no slower than ibnetdiscover in terms of sending and receiving
MADs. What you are likely seeing is some set transaction which is not
completing the way it should and this is being retried ad nauseum.
OpenSM is currently bad about this (I'm not sure it ever gives up).
ibnetdiscover gives up after 3 retries (and since it is only doing gets
this is less likely to be a problem). 

>  What kind of messaging could I do once i have a 
> network all worked out via ibnetdiscover? 

Not sure what you mean here by messaging.

> In other words, could I have a usage of infiniband that worked something 
> like this;
> - ibnetdiscover, save that info
> - message layer uses that info and does communications over ib

You can save that info with ibnetdiscover. What kind of messaging would
you want to do ? RC ? UD ? The message layer could utilize this info to
preclude some of the SA queries (not all if there are sets like
multicast or services or traps) but this subverts the SM/SA (and
replicates what it does). Also, if things change, this info would need
to be updated. In the case of this email, I'm pretty sure the SA is not
the issue.

> also, what's the next step to finding the problem here? OK, ibnetdiscover 
> works. opensm does not. What tool would you run next to look for problems?

The opensm log preferably in verbose (-V) mode.

Thanks.

-- Hal