Re: GOBACK (was: Perform Thru/Go to vs. Perform - Compile Speed

From: Michael Wojcik (mwojcik_at_newsguy.com)
Date: 05/18/04 

Date: 18 May 2004 21:17:08 GMT

In article <40a55e57.325467891@news.optonline.net>, robert.deletethis@wagner.net (Robert Wagner) writes:
> riplin@Azonic.co.nz (Richard) wrote:
> >robert.deletethis@wagner.net (Robert Wagner) wrote
>
> >> AS/400 had its genesis in a 1970 revamp of S/360 to something called NS
> >> (New System).

It was called FS, for Future Systems. FS was cancelled circa 1976.
Legendary IBMer Lynn Wheeler was partially responsible; his research
into the (abysmal) performance of FS was ammunition for those arguing
against it. His archives (at http://www.garlic.com/) have various
details.

FS was intended as the OS for either the 370 or a successor
tentatively named the S/380. It was partially intended as a clone-
killer, and partially a grandiose attempt to incorporate every
advanced OS feature anyone at IBM could dream up in the hope of
creating an OS so "advanced" it would drown all of its competitors.

> >No. AS/400 has nothing to do with 360. It comes from the S36, S38
> >line.
>
> You're wrong. It came from a failed Big Iron concept that found a home in
> S/36-38 land.

You're both right (and wrong). The AS/400 ("Silverlake") project was
the migration strategy for S/36 and S/38, and OS/400 included an
S/38-compatibility mode for many years (it may still, for all I
know). OS/400 incorporated many ideas from FS, for a variety of
reasons; one is that some of the (many) people who worked on FS went
on to work on the S/36 and S/38 (which were successors to the S/3;
the S/34 was a stripped-down S/38). Another was that hardware had
improved to the point where some of the FS concepts, such as very
rigorous supervision of application behavior, were more feasible
(though the original AS/400s were still painfully slow, particularly
the little development boxes).

So the AS/400 belongs to the S/38 family, but OS/400 is most closely
related to FS, which was intended as an OS for the S/370 family
(though never completed, much less released).

> >Actually AS/400s are Power machines too now aren't they ?

Yes, for several years now.

> Probably. Under the covers, nearly all modern machines are RISC emulating
> CISC.

While this is arguable in a certain limited sense (most contemporary
CPUs are 8-bit microcontrollers, after all, and it's debatable
whether "emulating" is the proper term for a RISC core processing a
microdecoded CISC instruction stream), this has little to do with
what kind of processor AS/400s use. The 400s are built on POWER
CPUs now because 1) IBM owns POWER, and 2) POWER has better
performance characteristics (particularly in terms of things like
power consumption, which are more important in data centers than
they are with home users) than the only other currently-available,
viable, comparable architecture, IA64. (The 400 uses the 64-bit
POWER designs.)

> Natively RISC machines are VERY fast, especially Sun.

RISC architectures aren't inherently fast. The best SPEC numbers
right now belong to x86-family CPUs, because Intel and AMD have
cranked their clocks up to ridiculous speeds. Those are hugely
inefficient chips, but economies of scale have made them more
affordable than any other architecture for most purposes.

There are applications where a super-hot, power-devouring CPU is not
the best choice, and that's where something like POWER finds its
niche. (Or where people want a decent 64-bit implementation. Or
where they want binary compatibility with another architecture
because they've invested in it.)

It's true that the modern RISC designs, from POWER on, are much
better (in terms of instructions per cycle and so forth) than the old
CISC designs. But since the high-performance CISC CPUs are now RISC
cores with a layer of CISC-instruction decoding around them, it's a
moot point.

What it really comes down to is that there are a number of good ideas
for creating very fast CPUs; everyone is using the ones they can
afford; and Intel (and AMD) is throwing a clunky compatibility
wrapper around theirs, because they have to in order to keep their
customer base, but the core is so fast that they still get the
performance they need.

And "especially Sun"? Sun's best SPARC performance numbers aren't
close to IBM's best POWER ones. They're not close to Sun's best x86
numbers, in fact. Take a look at http://www.spec.org/. SPARC was a
nice design in its day, but Sun doesn't have the resources to keep up
with chip design and fabrication innovation. They gave it a good try
and kept SPARC viable for an impressively long time, but they're not
Intel or IBM. Inventing a usable silicon-on-insulator process and
retooling your fab to use it, for example, is a huge investment. Sun
never had that kind of money.

> Both Sun and HP have
> announced they're abandoning their successful platforms in favor of Intel
> lookalikes. Visionaries know something we don't.

They know that it's cheaper for them to use Intel chips (and clones)
because of the aforementioned economies of scale. Sun does, anyway.
What HP is doing is anyone's guess, particularly since the Compaq
merger. And long before its decline HP got in bed with Intel for
IA64 - they decided VLIW was the way to go, and by partnering with
Intel they could keep the best of PA-RISC in a processor family that
would enjoy much larger volumes. As it turned out, of course, IA64
is a dog and almost no one wants it (though probably Intel will
eventually throw enough resources at it for it to succeed).

ObCOBOL: GOBACK. 

-- 
Michael Wojcik                  michael.wojcik@microfocus.com
This record comes with a coupon that wins you a trip around the world.
   -- Pizzicato Five

Relevant Pages

  • Re: Computer
    ... there were dualcore g4s and low power g5s ready to ... be used in powerbooks but apple had already decided to switch. ... Dual-core designs use a lot more power than single-core. ... The Intel dual-core designs have ...
    (rec.photo.digital)
  • Re: Computer
    ... future Intel designs. ... there were dualcore g4s and low power g5s ready to ... market-leading designs. ... already platform agnostic and ran on intel before it ran on powerpc ...
    (rec.photo.digital)
  • Re: RC Transmission Lines (Wafer-Scale)
    ... OSes that consume vast quantities of CPU power performing, e.g., ... e.g., AMD, VIA, etc. AMD was doing well for awhile with their ... Athlon CPUs, but Intel came back with cheap dual-core units and now ...
    (sci.electronics.design)
  • Re: Computer
    ... future Intel designs. ... market-leading designs. ... I/O is one place where speed costs power as the line drivers require ... a lot of Altivec fanbois out there who swore eternal damnation to Apple ...
    (rec.photo.digital)
  • Re: Intel Inside no more
    ... Aggregate meaning every machine out there, that a low power AMD could have done. ... Figure that there are roughly 200M PC processors sold each year, AMD has about 17% of the market, Intel has about 81% of the market. ... Given that we're referring only to Athlon64/Sempron vs. Pentium4/Celeron here, we're really only looking at the last two years worth of CPUs and only desktop CPUs. ... The above assumes that the computers are powered on ...
    (comp.sys.ibm.pc.hardware.chips)