🚀 go-pugleaf

sino wrote:
> 
> All,
> 
> I am a student in Computer Science.
> In this year, I need to study "Modern Operating System" in class.
> Now I need to prepare a article regarding "Why Cray supercomputer don't need
> implement Virtual Memory?"
> 
> Do u have any article mention that? Could u give me more information?
> Where to find those technical info?
> 
> All help is appricated.
> 
> Thanks
> 
> Sino
> sinolum@hk.super.net

Mr. Cray was asked this question by a stockholder at the 1979
annual meeting.  Mr Cray went through some general references to the
fact that he had always designed systems for raw speed & provided the
largest memory sets commercially available at the time of system
development.  

A later example of this was the Cray 2 with nearly a gigaword
(at 8 bytes to the word) of memory on the machine.  This machine was
available in 1984 & the largest disk systems available at the time
could not hold a gigaword of storage on a spindal!  "Besides," Mr.
Cray stated, "it takes another level of hardware (page fault & address
registers, et. al) to manage memory.  I have always striven for speed.
If you need the memory, then why not just have it on the machine?"
<Note that I an quoting from somewhat faulty human memeory here.>

One point to make here, Mr. Cray's systems were designed for a batch
environment where a job is loaded & run.  Constant context switching,
a la' vi-type interactive operations have always killed performance 
on these type systems.  VM-systems are ideal for interactive operation
because paging operates in a much smaller transfer environment than
swapping.  To swap a 10 g-byte job is ugly.  To page 2 m-bytes of
a 10 g-byte job is much easier on contention.  On the ther hand,
load the 10 g-byte job in & let it scream vs. the page operation
with the page resolution & fault hardware (did I forget to add the 
complexity of the os-loading) and you suffer in a batch environment 
with a VM-type system.

And, of course, for those who have dealt with Seymore jokes (an old
tradition in the company), when Mr. Cray was asked, "How come you 
didn't impliment Virtual Memory?"  He quipped, "' Cain't use what
you ain't got!"  In reality, the answer is as above.  The video tape
of this meeting may still exist, but I haven't seen it since 1986.

Oh, and I should also point out that stack hardware didn't exist
either.  Stacks were (reluctantly) implimented in software by a bunch
of ***programmers***!

One other point that you may want to note.  Mr. Cray designed the
Cray-1 hardware.  Programmers were only hired to write an OS 
because NCAR (the first paying customer) insisted that the system
come with an OS, or they weren't going to buy the machine!

Enjoy,

-- 
David Ecale
ecale@cray.com
"The difference between a wolf pup and a German Shepard pup is that a
wolf pup is quite happy teething on the leg of a stag that it's parents
brought down in a hunt while the German Shepard pup prefers to teeth 
on remote controls and high end graphing calculators...."