Re: MPI handlers need preemption

Dennis Cottel (dennis@rats-b.nosc.mil)
Thu, 9 May 96 11:21:16 PDT

Messages sorted by: [ date ][ thread ][ subject ][ author ]
Next message: Rolf Rabenseifner: "Re: draft of chapter 4 for next meeting -- pls comment"
Previous message: Marc Snir: "draft of chapter 4 for next meeting -- pls comment"
Maybe in reply to: Dennis Cottel: "MPI handlers need preemption"

> From: joel@ssd.intel.com
>
> -> From: dennis@rats-b.nosc.mil (Dennis Cottel)
> ->
> -> Since then, we've done some experiments, which Joel has confirmed, that
> -> show that Paragon NX handler code starts running within a few hundred
> -> microseconds. The handler does not wait for the computing process to
> -> voluntarily give up the CPU, nor does the handler wait for the computing
> -> process's time slice to be exhausted.
>
> I have to say I believe this "immediate" handler execution is an artifact
> of the unix-like Paragon scheduling and priority ageing and is not
> guaranteed by the functional definition.

I disagree with what I think is Joel's implication that immediate
handler execution is a sort of accident of implementation. Long before
the Paragon was announced, we were using NX handlers on our Intel
iPSC/860 to do things we couldn't do with regular (synchronous or
asynchronous) sends and receives because handlers provided immediate
execution. I claim that the implementation of the NX library on the
Paragon provides immediate handler execution because that is the
behavior that is needed and had come to be expected by Intel NX users.

Dennis Cottel, dennis@nosc.mil, (619) 553-1645
NCCOSC, RDT&E Division (NRaD), San Diego, CA

Next message: Rolf Rabenseifner: "Re: draft of chapter 4 for next meeting -- pls comment"
Previous message: Marc Snir: "draft of chapter 4 for next meeting -- pls comment"
Maybe in reply to: Dennis Cottel: "MPI handlers need preemption"