[openib-general] ucma into kernel.org

[Michael S. Tsirkin]

Quoting r. Or Gerlitz <ogerlitz at voltaire.com>:
> Subject: Re: ucma into kernel.org
> 
> Michael S. Tsirkin wrote:
> > So what are you saying? What did you learn from this system? What does this
> > say about CMA timeouts? That any timeout value is as good as any other? That
> > packets are never lost? This is the part that I am not getting.
> 
> I have learned that without tuning the CM timeouts/retries running
> N x m (m <= M) CM re-connection-ing in parallel worked fine (eg it was 
> part of the acceptance to kill a lustre server and have the all the 
> clients reconnect to the ghost of this server).

OK, so that's one application. But this does not tell us much unless we know how
gen1 CM managed timeouts as compared to gen2, does it?

> I can check the values if anyone can think it may be of help.

I guess we need this info if we are to learn anything from your setup.

-- 
MST