Very Computer

Please let me know what you know about the reliability
of the Iris Silicon Graphics workstations.  I heard that
the system reliability is very low via rumor posted on
INFO - IRIS.  Please let me know as reliability is an
issue and we may purchase shortly.

I don't read INFO - IRIS, sorry I can't scan your rumor.

When I worked at SGI the 3030 model (68020 w/ large disks) in my office
failed _very_ frequently, like twice a day - here failure is defined as a
_very_ reliable hardware _freak_ event, thought to be due to design errors
in the power subsystem, no, bogus disk controllers, no,... there was no
known answer before I left the company.

Most of the 10 or so new machines in the same area of the company (kernel
group) were equally eventfull. We all sighed and rebooted, hardware was
not our dept...

Please note that this was the view of a neo-dozen production machines of the
3030 model, all with a similar date of production, summer 1986, and all kept
with in a small portion of the 2nd floor of the engineering bldg of the
Stierlin Road site.  Maybe they got fixed, maybe you are referring to some
other model.

I would look elsewhere, of course, my reasons are not necessarily the same
as yours. I take into consideration the system software because that seems
to me as important as the window manager and the pixel painting primitives
Due to the SGI-only XNS dependency, and the TCP-on-a card of the 1986 IRIS,
it is a very hard machine to use in a distributed workstation environment,
where some of the workstations just happen to be very sharp color graphics
boxes.

I think that is what is important in my site's long term, and in just about
any hetrogeneous, open site, so I don't think about the IRIS seriously.
I'm making that point to the Navy - who wish to use it for a _very_ snazy
mailer system - they saw the SGI demos, not the network nor the "useability".
--
how about a great big spidery "X"?

The following views are my own.  I am not an official voice of SGI.

Quote:> When I worked at SGI the 3030 model (68020 w/ large disks) in my office
> failed _very_ frequently, like twice a day - here failure is defined as a
> _very_ reliable hardware _freak_ event, thought to be due to design errors
> in the power subsystem, no, bogus disk controllers, no,... there was no
> known answer before I left the company.
> Please note that this was the view of a neo-dozen production machines of the
> 3030 model, all with a similar date of production, summer 1986, and all kept

I have a production 3030 delivered to me in November '86 shortly after I
started working here.  I have had 2 hardware problems: a bad memory board
a week after delivery and a problem with the cartridge tape drive.  I have
had no hardware related crashes except those caused by parity errors in the
bad memory board prior to its replacement.

Here is the uptime(1) output for my Iris.
  2:51pm  up 11 days,  4:09,  7 users,  load average: 0.19, 0.24, 0.20

The last rebooted was for a new kernel.

Quote:> Due to the SGI-only XNS dependency, and the TCP-on-a card of the 1986 IRIS,
> it is a very hard machine to use in a distributed workstation environment,

> I think that is what is important in my site's long term, and in just about
> any hetrogeneous, open site, so I don't think about the IRIS seriously.
> I'm making that point to the Navy - who wish to use it for a _very_ snazy
> mailer system - they saw the SGI demos, not the network nor the "useability".
> --

My 3030 (Ramoth) has 4.3bsd's tcp code in its kernel and all of the 4.3
network utilities available.  I have no knowledge of the above
mentioned tcp on a card.  Ramoth is connected to a large network of
Irises, Suns and Vaxes.  and is also running nfs.  I'm using it to read
news right now.  I see no problem with interoperability.  Our famous
network dogfight program has even been converted to use tcp.  Enough
said?

The bottom line is find out the facts before making decisions.  Do not
rely on rumors and out of date memories and hearsay.
--
From the TARDIS of Mark Callow

"There is much virtue in a window.  It is to a human being as a frame is to
a painting, as a proscenium to a play.  It strongly defines its contents.

Quote:

>Due to the SGI-only XNS dependency, and the TCP-on-a card of the 1986 IRIS,
>it is a very hard machine to use in a distributed workstation environment,
>where some of the workstations just happen to be very sharp color graphics
>boxes.

Until the current (3.5) release, TCP was done via some code that was
downloaded into the Ethernet controller.  It was a dog - we were pushing
a lot of data through the ethernet, and the controller (or just a current
TCP connection) would hang quite regularly.  We actually had to have
a background process that would detect a hung connection and reboot the
machine in order to get it back.  It was slow too - about half the
throughput of XNS.  We eventually switched back to XNS, even though
getting SGI's binary distribution of their 4.2BSD XNS driver to work
with our modified 4.2 kernel was a lot of work.

The current version of TCP is in the UNIX kernel, not the Excelan one.
Also, SGI has switched to TCP as their standard protocol, making XNS
an option.