CUG.log Contents Back issues CUG SGI Contact us

N E W S   &   F E A T U R E S

CUG 98: A conference to remember

Corporate changes, marathon T90 BOF, the new SVI . . .

Barry Sharp
Barry Sharp,
regional director of the Americas


by Barry Sharp

The 40th CUG held in Stuttgart Germany was a great success, with attendance reaching 275. The local arrangements people were friendly and did a marvelous job running the conference smoothly for all of us. Many thanks should go to them for their time, effort, and smiling faces.

Mind you, the elevator voices were something else again ...!


Corporate changes

I sensed that many of us were uneasy with both Bob Ewald and Irene Qualters leaving July 1. Of course, it's difficult for any of us to know what really caused these two resignations. The hallway gossip lead me to believe that both Bo and Irene felt the company would be better if they left.

There seemed to be no rift between Bo and Irene and the new team of Rick Belluzzo, CEO, and John R. "Beau" Vrolyk, head of the Server and Supercomputing Business Unit.

Beau announced that most of Irene's duties will fall to Mike Dungworth, who has relocated from Mountain View to Eagan. Beau replaces Bob Ewald and appears to be a fast learner of the supercomputing jargon. Same is true with Rick Belluzzo.

There is no doubt that both Rick and Beau's first objective is to ensure that SGI returns to a profitable state. They both talked about high-performance computing and high-end graphics being their core products. However, they didn't spell out detailed plans using financial data to back up their comments about 1) where they are from a profitability perspective, 2) where they want to go, or 3) how they are going to get there.

I learned that SGI has been split into six business units, with the Supercomputing and Servers being one of them. They're supposed to get to a 10% profitability level as a unit. What if they don't? Will the Supercomputing and Servers Business Unit close down ? We'll need to monitor this aspect closely over next few years.

I speculate that SGI management will be in a state of flux for a while during the transition to the new business unit organization. This will surely have impacts on the SGI technical folks as well.

Denice Gibson (vice president, Strategic Software Organization) indicated she's thinking seriously about moving important long-term software development from Mountain View to Eagan, as attrition and/or turnover in Mountain View is bad; people can change companies without changing their parking spots! The service organization headed by Ken Coleman appears to have a low rate of attrition, probably the lowest in industry at around 2%. About 2,000 people report to Ken. So in this area, things look good for CUG member sites.


The T90 problem

I attended the T90 birds of a feather (BOF) session, which ran for nearly two hours! Vito Bongiorno, T90 product manager, announced that there have been 60 T90 installations since the product was announced: 29 T94s, 20 T916s, and 11 T932s. They have delivered some 551 CPUs (268 with Cray F.P.). Dave Kiefor stated that the mean time to interupt (MTTI) for T94s today is at around 2,000 hours, for T916s around 370 hours, and for T932s around 320 hours. As a comparison, MTTI on the Y-MP in its final years was around 2,500 hrs.

The main reason given for the T90 failures was the ECD (Emmitor Collector Dotted-logic) chip used in CPU, memory, and network modules. From around May '97 onwards this problem has become worse. It has caused the DOA (dead on arrival) rate to go to 10% (and much worse for some customers). Of these DOAs, they find that 50% of the returned modules are NTF (no trouble found). This is a very high ratio.

Dave showed a photograph of where the disconnect occurs on the wafer. The disconnect occurs where two dissimilar alloys join together along a planar surface. (Cray is responsible for connecting these wafers, while Motorola produces the wafers.)

Cray is now in the process of ensuring that these disconnects don't occur. It appears that they are close to the root causes of multiple problems with T90s. Based on what I heard in this BOF, I don't think the situation will be any better than it is for the rest of 1998. Improvements should be expected during 1999 for T90 reliability.

Several T90 sites gave miserable status reports. San Diego bought a new T916 with 14 CPUs in December 1997 and have replaced 17 CPUs since then. AEC in France has a T93 with 24 CPUs and they were down for ten days during April-May '98 due to network module chips going bad. Their long down time was made worse due to holiday schedules and the lack of expert help and parts in France.

Everybody said that T90 works like a champ when it is up!


The new SV1

S GI announced their new high-performance computer, the CRAY SV1 -- a big surprise for most of us at Stuttgart. First customer delivery was slated for August and general availability around December 1998. The SV1 has an attractive rebate for old computer equipment of up to 50% of its original purchase cost. This rebate offer is good until June 30, 1999.

Many of us asked SGI where the SV1 came from; it wasn't on their product roadmap six months ago.

The SV1, according to the new product roadmap, is a follow-on to the popular J90. I'm sure many of us will be exploring the SV1 to see if it fills any voids between now and the SV2 in 2002 (hopefully).

The SV1 runs UNICOS 10 and is compatible with CRAY J90s, YMPs, and C90s. Does this mean a T90 executable built with C90 mode can run without problems on an SV1 ?

Mike Booth talked enthusiastically about the SV1's multistreaming processor (MSP). This is where four physical CPUs are ganged together to appear as one CPU to the user. Today, the programmer must code for MSP, but tomorrow the compilers will take care of things! Hopefully the new SV1 customers will present their experiences to us at the upcoming CUGs.


A conference to remember

O verall, the Stuttgart CUG will be memorable with Bo and Irene leaving, new SGI management showing their colors, a large attendance, a well-run conference, a new Special Interest Group (SIG) structure, new faces, good presentations, public transportation, and excellent hosts.

The museum visit on Wednesday night was also of great interest with its tasty food, many interesting people to talk with, the museum shop, and more than 100 years of automotive history surrounding all of us. Sally Haerer, our new CUG president, was wheeled around the museum on three wheels in style by the curator.

rule
Contents      Back issues      CUG      SGI      Contact us