PCs vs. old minis and mainframes (was Re: gauging interest in VAX 6000-530)

From: Eric Smith <eric_at_brouhaha.com>
Date: Mon Oct 25 02:16:02 1999

Allison J Parent wrote:
> I think FDDI is still in the 100++m/bytes/Sec region.

No, FDDI is about 100 megabits per second. And it is fairly complex,
typically requiring more host processing than Ethernet, so usually a cheap
100-base-T card on a PC will match or outperform the throughput. Only
in very contrived cases that rarely (if every) occur in the real world
is FDDI a win. So why was FDDI around at all? Simply because 100
megabit Ethernet hadn't yet been developed.

> I will not argue this as PC are finally going fibre and Gigabit eithernet
> but thats 10 years after the fact.

FDDI cards were available for PC at about the same time as they
were for VAXen, so this certainly doesn't help prove that a VAX can
outperform a PC.

Gigabit ethernet cards for PCs actually showed up *before* such cards for
minicomputers. In fact, this phenomena is exactly why most minicomputers
now support PCI.

> Though there were a few parallel busses that were 50-100mbytes/sec rate.

Backplane busses yes. But not any common peripheral or network interfaces
that were used on VAXen in the late 1980s.

> Two general catagories, memory interconnect and Storage interconnect.
> There are several IO interconnects as well.

Just as most VAXen had a memory interconnect that was faster than their
I/O and network interfaces, so did the 486 PCs. This argument is another
red herring that does not prove that a VAX outperforms a PC.

As far as peripheral and network interfaces go, almost all of the very
high-speed stuff is serial (electrical or optical), with the seriously
high-speed stuff using parallel optical.

Personally I'm getting tired of the type of rants about how PCs have such
terrible I/O bandwidth and that any old 360/30 from 1965 could whup their ass
because the 360 had I/O channels.

Fact is, the cheapest PCs you can buy today have better I/O bandwidth to
disks and networks than all but the highest performance mainframes of
the 1980s.

So why are modern PCs barely adequate to serve a single user, whereas
1960s-era mainframes that were hundreds of times slower could serve hundreds
of users?

Simple. The type of software that was run on the mainframes was MUCH
different than what we run on PCs.

If you ported the OS and applications from a 1960s mainframe to a
suitably equipped modern PC (and I'm not talking about a $500 special, but
a PC with a serious disk system), the PC would outperform it by the
hundreds of times that you would expect.

Similarly, as Mike Cheponis pointed out, a good 486 box WILL outperform
a VAX 8650, if you run equivalent software on it. The 486 box has not
just more CPU, but also more I/O bandwidth.

Note that Mike did NOT say that about the VAX 6000. He was comparing to
the 8650 because the benchmark figures were handy.

Mark Green <mark_at_cs.ualberta.ca> wrote:
> Time for a short lecture in hardware architecture (you can tell
> where I'm from :-)). Integer performance is a very misleading
> measure of performance when you are talking about system
> performance.

Yes, that's true, but when the difference is very large, you can't just
handwave it away by claiming that benchmarks are no good. Even if the
dhrystone benchmark totally sucks, the fact that the 486DX2/66
outperforms the VAX 8650 by a factor of five is too large a difference
to sweep under the carpet by saying that the VAX isn't good at
dhrystones.

Personally, the reason I have negligible interest in a VAX 6000 has
nothing to do with performance. If I was concerned about absolute
performance, I wouldn't be collecting PDP-8s and 11s. In my view,
almost all VAXen but the 780 and 785 are completely uninteresting
because they are full of custom parts that can neither be repaired if
they break, or studied and understood.
Received on Mon Oct 25 1999 - 02:16:02 BST