linux-mips
[Top] [All Lists]

Re: mpp kernel interface

To: lm@gate1-neteng.engr.sgi.com (Larry McVoy)
Subject: Re: mpp kernel interface
From: Alan Cox <alan@cymru.net>
Date: Thu, 16 May 1996 10:09:17 +0100 (BST)
Cc: tridge@cs.anu.edu.au, linux-mc@arvidsjaur.anu.edu.au, Linus.Torvalds@cs.Helsinki.FI, linux@neteng.engr.sgi.com, alan@cymru.net
In-reply-to: <199605152243.PAA19394@neteng.engr.sgi.com> from "Larry McVoy" at May 15, 96 03:43:19 pm
Sender: owner-linux@cthulhu.engr.sgi.com
> : With 30 people now on this list there must be someone else who wants
> : to say something ...

I've not actually seen anything from the list (should be going to ukuu)

>       SMP vs cluster?

Why vs, why not AND. An SMP box is just a couple of well connected nodes.

> The hard one is sockets.  I've never seen a good solution to that.
> I'll come back to that.  First, I want to go through the other ones and
> offer suggestions.

There are several approaches to this. One is to treat it like a device so
you open a socket on a given machine and it like the device doesnt walk. The
other is to bend IP mobile to the needs. That would probably want something
like an old 486 as a network FEP.

> do remote processes.  For the PID name space, make pid_t a 32 bit int,
> make the top 16 bits the host part, and the bottom 16 bits the pid part.
> (We need to come back to the host part when we discuss process migration.)
> A host part of "0" means "this host".  So a "kill -HUP 1" will always
> restart init.

We can also do this for devices so you can mknod a printer on a different 
node.

> Devices I sort of punt on.  For device access, I would just use the 
> remote mag tape protocol (or something very, very similar) so that all
> of the locking, etc., still works - since you ship all the requests to
> the system w/ the device, that kernel can implement the locks.  Any
> issues here?

The vfs issues fairly controlled requests to the FS layer, and the device
layer also is fairly clean. The MOSIX project intercepted stuff at this
level --- so a remote device turns the request into a message. The system
also accounted messages so a process like a find would migrate across cpus
as it changed the disk it was searching.

> we have the talent right here on this list to do it.  So I'll bow out of
> commenting on it, other than to say make sure mmap works right.

mmap is foul. SYS5 shared memory is just as bad too. 

> Sockets.  This is a hard problem.  Some people think that a socket
> should stick around after the CPU that created the socket has crashed.

If you are having a single apparent IP address this is true for TCP, you can
on spotting a down host just send an RST and go into TIME_WAIT to preserve
the corruption protection properties of TCP.

> is a big performance win.  Coming from a network cluster, you'll get that
> without having to work for it - the other way frequently is harder.

It also means more people can play


<Prev in Thread] Current Thread [Next in Thread>