Xref: utzoo comp.sys.next:1148 comp.misc:4596
Path: utzoo!attcan!uunet!lll-winken!lll-lcc!ames!mailrus!eecae!cps3xx!rang
From: rang@cpsin3.cps.msu.edu (Anton Rang)
Newsgroups: comp.sys.next,comp.misc
Subject: Re: NeXT and sources (really "Do we need OS source?")
Message-ID: <1438@cps3xx.UUCP>
Date: 5 Jan 89 18:25:40 GMT
References: <8812220917.AA10582@decwrl.dec.com> <29954@tut.cis.ohio-state.edu>
Sender: usenet@cps3xx.UUCP
Reply-To: rang@cpswh.cps.msu.edu (Anton Rang)
Followup-To: comp.misc
Organization: Michigan State University, Computer Science Dept.
Lines: 76
In-reply-to: verber@cheops.cis.ohio-state.edu's message of 22 Dec 88 16:45:50 GMT

In article <29954@tut.cis.ohio-state.edu>, Mark Verber
 (verber@cheops.cis.ohio-state.edu) writes:
>In article kevin@hiatus.dec.com (Kevin Baranski-Walker) writes:
>>> I stressed to him the importance of sources, both for system administration
>>> and for research ...
>>
>>Gee I got alot of work accomplished w/o MVS, RT-11, or VMS sources (pre-DEC
>>of course :-)

>There can be a lot of arguments about source/no-source.  Almost all
>sites *think* that they do need source.

  I agree with this--most people think they want source for the OS.
They probably don't need it, though.

>I agree that many sites don't need source,
>but saying that sources to the OS aren't needed is just as extreme as
>saying that everyone needs source.

  So make the source optional (as DEC has done, I believe).  That way
those of us who DON'T need source don't wind up paying for it.

>There are other sites that this just doesn't work.  These sites would
                            ^^^^ [not having source]
>be characterized by one of three things: size, security concerns, or
>computer science research.
> [ ... ]
>There are many sites out there that have a large collection of hosts
>of varying types.  These sites often want the systems to have access
>to identical services and resources.  Sometimes these sites have need
>to drop things into the kernel.

  A decent OS should let you write kernel-level code *WITHOUT* needing
to modify the OS itself.  For instance, VMS lets you write device
drivers (where I/O is involved) and user system services (for general
types of things).  The interfaces are documented.  No source needed
(though it can help by using parts as an example).
  Having sites modify their OS is a vendor's (support) nightmare.  If
a large number of changes are being made (say, implementing NFS under
VMS :-) the chances that the customer will be willing to understand
all of the relevant code (some 250 microfiche sheets for VMS/RMS) are
pretty low.  And what about upgrades?

>Lets face it, all software vendors are slow when it comes to
>distributing fixes.  This is a fact of life right now.  What happens
>if a major hole is found in the NeXT OS at a secure site?

  This can be a problem, but having source isn't a "magic" solution.
The larger, business-style vendors (DEC and IBM, say) generally are
quite good about security patches (we reported one major VMS problem
and got a patch tape back within a week).
  If we had the source, maybe we could hack around and fix the
problem.  Maybe we could have broken the three or four modules which
depended (in non-obvious ways) on various parts of its behavior, too.

>I work for a computer science department.  My users are researchers.
>One research project is investigating system preformance.  Another
>project is looking into different ways to do distributed systems.

  For this kind of thing, you really do generally need source.  Why?
Because you're writing an operating system.  It's not a question of
"modifying" the existing OS as much as making it do things it wasn't
intended to do--adding paradigms.  Of course, you then wind up with a
myriad of differing versions, but that's OK for research.

  Just my opinions.

>Mark A. Verber
>Ohio State Univ.

	Anton

+---------------------------+------------------------+----------------------+
| Anton Rang (grad student) | "VMS Forever!"         | "Do worry...be SAD!" |
| Michigan State University | rang@cpswh.cps.msu.edu |                      |
+---------------------------+------------------------+----------------------+