[28] in 1993-clients
Re: I've played with both and...
daemon@ATHENA.MIT.EDU (akajerry@Athena.MIT.EDU)
Wed Mar 11 17:27:56 1992
From: akajerry@Athena.MIT.EDU
Date: Wed, 11 Mar 92 17:27:27 -0500
To: Bill Cattey <wdc@Athena.MIT.EDU>
Cc: 1993-clients@Athena.MIT.EDU
In-Reply-To: [19]
> I think a relevant thing to learn here is: Yes both the DEC 5000/25 and
> the IBM RS/6000-200 have different operating systems, but the DEC one
> has few enough differences that all our code just works. Not so the IBM.
Don't be afraid of the dark, Bill.
Athena has to wean itself off of BSD eventually. System V isn't bad,
it's just different. If Athena conitues to avoid System V then you
risk going the same way as DOS, backward compatable with something
that was a good idea at the time, but just doesn't cut it in the 20th
century.
Not that BSD is a bad thing, but just how many packages are there that
continue to be a pain to port because they rely on BSD style code, and
in some cases, even out of date BSD style code. The normal solution
has alway been to rely on the vendors BSD compatability libraries,
which in general is just asking for poor performance and bugs.
I've programmed on the RS/6000 enought to say I like it as a
development environment. Most of the BSDism I really like are there
and for those that aren't I don't really mind the alternative, in some
cases I prefer then to the BSDisms.
Also, I'll let you in on a little secret there are some people, even
at MIT, who prefer System V to BSD. Also, there are packages out
there that don't port so well to BSD.
> The downside is that there isn't a REAL Athena integration, and that we
> run the risk of having a fiasco like the initial deployment of the
> RS/6000 and the lesser mess on the initial deployment of the DECMIPS
> machines.
>
> But as the most vocal critic of premature deployment of new vendor
> platforms, I say that the Sun represents the opposite situation. It
> think it would be a bigger mistake to deploy the RS/6000 with its
> plethora of little operating system differences that are STILL jumping
> out at us and its messed up font integration, or to settle for a the
> blurry lower resolution color screen of the DEC 5000/25 than to go with
> the Sun platforms.
>
> The preliminary Athena integration has turned up fewer snags than the
> DECMIPS, RS/6000 and possibly even the RT integrations.
>
> The Sun operating system has been out in the world being debugged for as
> long as our BSD operating system. This represents a unique opportunity:
> We won't have to debug the vendor's operating system for them! (NOTE:
> this consideration is invalidated if Solaris 2.0 is taken instead of
> SunOS 4.1.)
Never underestimate the ability of Athena to pull a fiasco out of its
hat.
I haven't played with the Sun port much so I can't comment directly,
but I have dealt with enough ports (DECMIPS, AUX 2.0, AIX 1.2, AIX
3.1, half of which failed, I might add) to comment in general. The
Sun port has had the advantage of being an unofficial port and I'm
wouldn't be ready to call it a real port until the "managers" have had
their licks at it. Ports don't fail because the OS can't support an
Athena environment, they fail because the OS / hardware combination
doesn't fit the niche dreamed up for it by the managers, and its not
until the port is almost complete that someone asks, "is there really
a customer for this."
> It is clear that we will eventually deploy Suns as public workstations.
> But I believe there is a real opportunity cost to viewing the Sun port
> as being difficult: The RS/6000 port is still, in some senses,
> unfinished. The present, experimental, Sun port is already more
> compatible with the Athena computing environment than the AIX port!
> There is a real risk we'll compromise ourselves into being saddled for
> four years with an overpriced DEC solution that simply isn't up to snuff.
This sound like the exact opposite of the problem that doomed the AIX
1.2 and AUX 2.0 ports. There are really big niches out there
screaming to be filled and Athena refuses to fill it. Is there a need
for a Sun port? Hell yes! Is there a need for an RS/6000 port? Well
that's a little harder to answer, but I will say this. Try to find a
public cluster RS/6000 free after 11AM or before midnight, if the port
is that unfinished, no one seems to mind all that much. Also, I've
noticed alot of people using the RS/6000's for the real serious number
crunching that it was designed for, a niche in which it has the
advantage over everyone except HP.
One last note for the world of AI and genetics. If you ever play with
genetic algorithms in AI, you'll learn that natural selection isn't
just a matter of survival of the fitest, sometimes its also survival of
the most diverse.
--Jerry
