Michael - I think you've laid out a pretty accurate case here. I have the following comments.
On the issue of NT vs. Unix, a typical big site might have 1 UNIX servers for every 50 workstations. People run the core business apps on the UNIX workstations but also have Wintel on their desks for word processing, some DTP, spreadsheets, etc. At the ludicrous limit, thats two workstations for every man, woman and child in the organization! I'm seeing orgs replace the two boxes with a single NT box through attrition and using the X-emulators to get to the core apps on the Unix server. So I believe the biggest threat on the Unix boxes is on the workstation side, not the server side. I've heard similar issues raised by the admin folks I know but there are workarounds here and the end result is that the users are happy campers.
> Customers spent their IT budgets downsizing to desktop computers for
> non-critical tasks. Very little money was left for purchasing new
> mainframes which the customer did not need anyway because... As the
> non-critical tasks were off loaded from the mainframe, it actually
> increased the useful life of the mainframe for those critical tasks
> it still had to perform.
One major point here is that main frames do not have built-in obsolescense like every other piece of computer machinery. One of my customers just bought a new IBM Main Frame (huge mother!) to replace a machine that was 18 years old! They needed a bit more speed. Thats a pretty typical main frame life cycle. They truely are upgradable. I contend that IBMs problem is that everyone built in obsolesence and they didn't.
I disagree with your assertion that non-critical tasks were moved off of main frames. It was the other way around. Those tasks which were heavily transaction oriented were left on the main frame. Usually the "bread and butter" tasks. I don't know of a reputable bank or insurance company that would dare put their customer base on a client/server system and I work with a number of them.
> It might take only 1 million textured polygons per second to
> provide a realistic image to one screen. But it takes 20M polys/sec
> to drive twenty displays with different content or different
> perspectives on the same content. The question then becomes, what
> is the cheapest, most appropriate solution for this application?
> Buy twenty networked PC's using a distributed application. Or, buy
> twenty terminals and one SGI server using a client/server
> application.
At this time, SGI is not cost effective in this market. If you want a 3-channel system, you can buy 3 boardsets for an NT box (like NetPower or Intergraph) and get a decent system for $15-20K. To get any similar solution with SGI, you need to buy and Onyx for about $100K. Granted the performance is much higher but its not cost effective for entertainment. One entertainment company I know of tried selling this a couple of years ago and lost their shirt! I think SGI could do it but they don't have the will to do it at this point. This is one of the big reasons they are losing customers. I know its why two of my loyal SGI customers left!
Jeff |
