Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!ll-xn!ames!sdcsvax!darrell
From: darrell@sdcsvax.UCSD.EDU (Darrell Long)
Newsgroups: comp.os.research
Subject: Re: (Belated) Shared Libraries
Message-ID: <3163@sdcsvax.UCSD.EDU>
Date: Wed, 13-May-87 16:05:26 EDT
Article-I.D.: sdcsvax.3163
Posted: Wed May 13 16:05:26 1987
Date-Received: Sat, 16-May-87 11:43:39 EDT
Lines: 47
Approved: mod-os@sdcsvax.uucp

Hi,
	I would have thought that the best approach would be to use
something of the server idea. That is that each processor that needs to has
a server providing any necessary shared libraries, and the calls to them
could be replaced with a very simple RPC.  Since these ``servers'' will
occur on the same processor as the client the mechanisms necessary will be
simplistic.  To a large extent this is the same solution as having
libraries sitting in each machine and having the machine then tell the
process where the library is, but makes this problem the same as supporting
servers. This should reduce the size/complexity of the kernel/support
software.

	Similarily not all libraries need be present, but when necessary
they could be loaded, and only those that are needed be loaded, much as
with the 4.[23] BSD UNIX/SUN UNIX RPC server server.

	This would reply heavily on efficent IPC on a single processor in
order to be efficiently implemented, but I guess could be perfectly
feasible.  The most difficult areas would be, I suspect, firstly performing 
library calls that need part of what is usually the processes private 
information - things such as file descriptors or global variables (errno). 
And secondly when migrating processes certain IPC ``links'' would have to
be updated to point to the new machine and/or not changed at all
(connections for input/output vs local shared libraries).

	Given an environment where you are deliberately trying to exploit
parallelism I don't think that comparing libraries to processes is invalid.
It also allows libraries that call heavily on certain resources to be
appropraitely placed (i.e. vector libraries placed on a processor with
vector processing hardware).  Any thoughs?

				Kevin

---
  "I'm going to live forever, or die trying"
	      -- Vila, ??, Blake's Seven

  murray@uk.ac.york.minster or ..!seismo!mcvax!ukc!minster!murray  (I think)




-- 
Darrell Long
Department of Computer Science & Engineering, UC San Diego, La Jolla CA 92093
ARPA: Darrell@Beowulf.UCSD.EDU  UUCP: darrell@sdcsvax.uucp
Operating Systems submissions to: mod-os@sdcsvax.uucp