Mainframes vs PCs (was Re: discrete transistors)

From: Chuck McManis <cmcmanis_at_freegate.com>
Date: Mon Oct 19 17:32:06 1998

Ok folks, its true that mainframes typically have had better I/O bandwidth
to their disk farms but that has changed. The PCI bus can and does sustain
greater than 100 Mb/s over it. The "old" PCI bus spec had it at 33Mhz *
4bytes/write for 120Mbs and a properly designed PC can run the bus at 66Mhz
these days giving an opportunity for 240Mbs. AGP does somewhat better.

Now ISA does suck dead gophers through a garden hose, and that was why we
saw VLB, MicroChannel, and PCI.

Another thing that mainframes use that PC's typically didn't were channel
controllers. Those dedicated CPUs that sit on the Memory bus and blit
things into and out of memory rather than bothering the CPU with all that.
To facilitate this the mainframes typically have fairly complex snooping
caches for effective management of paging activity.

The Crays and ConnectionMachines have, in the past, had the advantage of
being vector processors where typical mainframes were often SIMD machines
at best and simple pipelines at worst. Microprocessors caught up with the
SIMD wave with multi-ALU pipelining, and with the Katmai and AMD-K7 they
will get many of the vector features that made so-called "super computers"
so fast.

If you build a "PC" (Pentium II class) with 256MB of SDRAM and dual PCI
based fast/wide SCSI controllers running to a striped RAID array of "good"
SCSI disks you can "beat" a lot of mainframes. Of course you best them with
a $10,000 PC.

Back to classic computers, it has been said, perhaps apocryphylly(sp?),
that "My laptop has more computer power than NASA used to put men on the
moon." While it may be true, I've never actually seen a description of the
computer resources available to NASA between 1962 and 1969. Does anyone on
the list have that information?

--Chuck
Received on Mon Oct 19 1998 - 17:32:06 BST