>Date: Wed, 10 Apr 96 17:13:13 met
>Reply-To: mklinux-powermac@gr.osf.org
>Originator: mklinux-powermac@gr.osf.org
>Sender: mklinux-powermac@gr.osf.org
>Precedence: bulk
>From: "Michael J. Prevost" <prevost@world.std.com>
>To: nocturno@i-2000.com
>Subject: Re: memory needs...and benchmarks....
>X-Comment: MKLINUX POWERMAC List
>
>>> What is MP?
>>
>> Multi Processor
>
>There seems to be a real mix of "old hands" and newbies on this list,
>so at the risk of boring the old hands I thought that I'd add to this
>answer a bit...
>
>There are two general classes of multiprocessor systems; loosely
>coupled and tightly coupled.
>
>Loosely coupled multiprocessors work almost as independent computer
>systems. Tasks (processes) have limited ability to coordinate work
>with each other. Separate kernels operate each processor in the
>system. The Radius and Daystar MP Mac's are architected this way. Each
>CPU runs a MacOS and applications are launched manually on a specific
>CPU. A complex task/application cannot utilize multiple CPU's
>simulataneously.
>
>Tightly coupled multiprocessor systems on the other hand are designed
>so that the CPU's can be utilized cooperatively. There are two main
>flavors of tightly coupled systems Massively Parallel (MPP) and
>Symetric Multiprocessor (SMP).
>
>OSF has done a tremendous amount of top quality work on Mach to add
>SMP capabilities to the microkernel. They are also adding a full suite
>of real-time features as well. Under SMP there is (usually) one "full"
>kernel that controls key functions (like booting up the system, many
>or most I/O, system clock(s) and timers, task scheduling,...). A
>partial kernel operates on each secondary CPU.
>
>Under a good SMP implementation (like OSF's Mach) a single process can
>be "threaded" so that multiple processors can be used to get the work
>done. There are libraries of thread routines to create, destroy,
>dispatch, coordinate (semiphores and the like), communicate (message
>passing, non-thread-private data structures and the like) threads.
>
>We are very, very lucky that OSF's Mach is the choosen microkernel
>because it gives us the ability (or at least the path to get) Posix
>thread libraries, real-time applications, and scalable SMP! This is a
>Linux that can challenge the likes of Sequent, Sun, Digital, Harris,
>Concurrent...
>
>We are fortunate indeed. Most of us should take some time to cruise
>OSF's web site in detail. You will like what you see.
>
>Sorry for the long ramble :-P
>
>Mike Prevost
>Dot4, Inc.
>