Path: utzoo!attcan!uunet!seismo!sundc!pitstop!sun!amdcad!ames!oliveb!tymix!antares!jms
From: jms@antares.UUCP (joe smith)
Newsgroups: comp.os.misc
Subject: Re: Multi-CPU system task scheduling
Summary: BiiN does it in hardware.
Message-ID: <267@antares.UUCP>
Date: 17 Nov 88 10:03:50 GMT
References: <2174@ficc.uu.net> <1125@igloo.UUCP>
Reply-To: jms@antares.UUCP (joe smith)
Organization: Tymnet QSATS, San Jose CA
Lines: 28

In article <2174@ficc.uu.net>, walker@ficc.uu.net (Walker Mangum) writes:
> 
> I'm working on a real-time multi-CPU system in which a process can
> be scheduled by another process, a timer, or an event in the "most
> attractive" CPU.  The actual mechanics of loading and scheduling
> the new process work just fine.  The problem is the definition of
> "most attractive" CPU.

This information may not help in your particular problem, but you might
want look at how BiiN is doing it.  Their CPUs have the ability to do
a context switch and pull the next task block off of a queue in a single
instruction.  The scheduler software builds a queue of runnable processes
that the hardware understands.  Whenever a CPU determines that it cannot
run the current process anymore (waiting for a resource, for I/O, or time
quanta expired), it picks the next available process to run.  There is no
need to schedule a job for a particular CPU; they are all identical.

In order for this to work, all I/O controllers have to be accessable to
all CPUs.  All CPUs see the same set of interrupts.  Whichever CPU that
can handle it first takes care of the interrupt.

I doubt that your system has this capability, but I find the concept
intriguing.
-- 
+----------------------------------------------------------------------------+
| TYMNET:JMS@F29  CA:"POPJ P,"  UUCP:{ames|pyramid}oliveb!tymix!antares!jms  |
| INTERNET: (Office-1.ARPA is no more)      PHONE:Joe Smith @ (408)922-6220  |
+----------------------------------------------------------------------------+