[ofa-general] Re: Re: [PATCH RFC] sharing userspace IB objects

[Gleb Natapov]

On Tue, Jun 26, 2007 at 10:35:12PM +0300, Michael S. Tsirkin wrote:
> > > applications *that's what currently happens*. If multiple threads poll a CQ,
> > > which one gets which completion is currently unspecified. Are you
> > > worried about this? If not, why are you worried when multiple
> > > processes do this?
> > You've missed my sentence about difference between multithreaded
> > application and what you propose. The difference is HUGE (I can't write
> > bigger letters sorry about that). I can design a multithreaded MPI so
> > that each thread will be capable to progress MPI send/recv request (and then
> > I don't care what thread gets which completion. I can't do it with multiprocess
> > scenario.
> 
> Well, with shared memory, the difference between thread and process is not that huge.
> And with the proposed API, you will be able to do just that.
> 
With your logic kernel can send signal to any process no matter which
process actually caused it. After all this is what it does with threads.
You are thinking about syntactic benchmark that just send random data to
a peer and free it on completion. The real program has much more state
associated with each operation and corespondent completion. And received
data have to be actually processed by a process it was send to and not
just by any process. Unless you'll stop repeating your mantra that
threads are just like processes with shared memory segment we will not be able
to address shortcomings of your proposal.

--
			Gleb.