Memory, bandwidth, and aliens

Or, CUG is a many-splendored thing

by Lynda Lester

T here is a great hall with seven chandeliers. Leaves twine through the carpet in Celtic designs, and long straight tables march across the room in rows, side by side, front to back. Three hundred people sit at the tables in chrome chairs, sipping water from stemware glasses.
At the front of the room, complex patterns move on a giant screen. "Once we have an Infinite Reality Engine, we'll have no problem visualizing this," says a voice over the sound system, which at that moment begins to skroink and buzz into cadences Rod Serling never imagined.

Aliens have commandeered the PA--again. I know it must be them because I have seen their ship, notably similar to certain vessels of Borg design, which has crashed outside across the street in front of the bank.
They bothered me yesterday when I gave my own talk in a corner of this very room, which had been sectioned into salons for special-interest presentations. I opened my mouth to say, "How do we entice users away from the comforting realm of retroprint into the new world of the wide web," but what came out over the mike was, "This is a systolic implementation of a dynamic programming algorithm."
For ten minutes, the voices of speakers in all the salons were scrambled through the PA as the audience tried to figure out How to Utilize Applications and Algorithms in User Friendly Newsletters by Networking for Performance and Evaluation on J90 Mass Storage Systems.
The aliens laughed. They are laughing now.
But this is the CUG zone. It is October, it is Charlotte, it is the Fall '96 meeting of the Cray User Group, and everyone is thinking this is the greatest conference since sliced bread--aliens or no.
300 high-performance intelligences are in phase; 300 turbo geeks are communing in a single language. The atmosphere is dense, condensed, intense. The jokes are Dilbert-like: "I said metarouters, not gigarings. Ha! Ha!" It's surprisingly comforting; after all, no one understands your job and the issues you face till you come to a meeting like this.
Per which a hapless Silicon Graphics, Inc. (SGI) man is nearly lynched after publicly suggesting that system administrators wanting source code couldn't read it even if they had it--a lack of awareness typical of what site reps may face in the Greater Outlands, but do not expect to face at CUG. CUG is something else: a place to grind grits and gnash nits, where users talk and vendors listen and vice versa, where everyone respects everyone else and a lot of work gets done.
Of course the epochal SGI/Cray Research merger did occur, and everyone is waiting to see if SGI will behave in a ravenous and Hunlike manner toward Cray, and if CUG will go to the dogs--but the outlook looks good, at least in beta.
SGI chief executive officer Ed McCracken deems CUG important enough to personally attend. He addresses the members of CUG in a plenary session, where he is attentive to the organization and emphatically optimistic about the SGI/Cray future. His remarks show not only confidence but humor and even a degree of humility--which, if it is PR, is good PR. "I got no respect till I put 'CRAY' on my business cards," he quips. "People used to treat me like a salesman; now they treat me like a colleague."
SGI's responsiveness to their new stepchild CUG is also embodied by Mark Goldman, who flies in heroically from SGI on a moment's notice to fill a program slot that has just opened. Goldman, manager of the Accelerated Strategic Computing Initiative (ASCI) project, tells the inside tale of how the newly merged SGI/Cray staff, eyeing each other with sweaty palms and trepidation, did a mind meld to win the Department of Energy's award for building a teraflops computer. (The ASCI concept: To stop bombs from blowing to bits in nuclear tests by turning bombs into bits inside computer models.)
Meanwhile, Bo Ewald, president of Cray Research, Inc. (CRI) and senior vice president of SGI, is a strong presence. He attends CUG planning meetings and addresses two plenaries, talking straightforwardly about challenges and opportunities for Cray. He also attends the CUG receptions, where he mingles with delegates and welcomes first-time attendees with a graciousness and respect toward users that is touching. If it is PR, it is excellent PR.
Vice president of CRI Supercomputing Irene Qualters steps up to the board and gives a presentation on the supercomputing roadmap. But first she tells us how she was riding her bike the day before and thinking about Seymour Cray's death, the purpose of life, and What It All Means. We feel like crying in our beer, then changing the world.
Cray Research also missions a cadre of hardware and software and service people who give talks, offer help, answer questions, dispense advice, and tell how they were stung by bees on the golf course.
rule
Outside in downtown Charlotte the air is sweet and humid. Historical plaques line the walks; art sculptures and commemorative statues define every corner. Fountains are omnipresent--enormous cascading waterfalls, pools with ripples and dancing waves. At sunset, gleaming glass buildings reflect a fiery sky.
It's good to get out once and a while. You have to sometimes, at least to escape the hotel coffee for a cappuccino next door.
CUG is great. It gives you these weird opportunities. Like, everyone has seen Los Angeles, but who's ever been to Charlotte?
rule
Back in the business meeting, people wonder if, with the SGI/Cray merger, we should change the name of the user group; "SiliCUG" is suggested. Debates rage over bylaws and amendments; new officers take charge.
The technical program continues. Presenters describe express-link options, source-synchronous links, and modular configurations. The audience asks about memory-read latencies and bisection bandwidths.
Memory, bandwidth, aliens--it's been real.
On the last day, David Robertson, local arrangements chair for the next CUG, invites us all to the San Jose meeting in May 1997. He promises that it will be an earthshaking event. "Silicon Valley is the epicenter of high technology, and this CUG will feature the latest upheavals in supercomputing. Leave your typhoons, tornadoes, and hurricanes behind and come to California for 'Seismic Supercomputing'!
"Don't let the theme scare you away," he adds. "There hasn't been an earthquake in months."

"I said
metarouters,
not gigarings.
Ha! Ha!"

We feel
like crying
in our
beer, then
changing
the world.

Contents || CUG home

Comments to: lester@ucar.edu

by Lynda Lester	T here is a great hall with seven chandeliers. Leaves twine through the carpet in Celtic designs, and long straight tables march across the room in rows, side by side, front to back. Three hundred people sit at the tables in chrome chairs, sipping water from stemware glasses. At the front of the room, complex patterns move on a giant screen. "Once we have an Infinite Reality Engine, we'll have no problem visualizing this," says a voice over the sound system, which at that moment begins to skroink and buzz into cadences Rod Serling never imagined. Aliens have commandeered the PA--again. I know it must be them because I have seen their ship, notably similar to certain vessels of Borg design, which has crashed outside across the street in front of the bank. They bothered me yesterday when I gave my own talk in a corner of this very room, which had been sectioned into salons for special-interest presentations. I opened my mouth to say, "How do we entice users away from the comforting realm of retroprint into the new world of the wide web," but what came out over the mike was, "This is a systolic implementation of a dynamic programming algorithm." For ten minutes, the voices of speakers in all the salons were scrambled through the PA as the audience tried to figure out How to Utilize Applications and Algorithms in User Friendly Newsletters by Networking for Performance and Evaluation on J90 Mass Storage Systems. The aliens laughed. They are laughing now. But this is the CUG zone. It is October, it is Charlotte, it is the Fall '96 meeting of the Cray User Group, and everyone is thinking this is the greatest conference since sliced bread--aliens or no. 300 high-performance intelligences are in phase; 300 turbo geeks are communing in a single language. The atmosphere is dense, condensed, intense. The jokes are Dilbert-like: "I said metarouters, not gigarings. Ha! Ha!" It's surprisingly comforting; after all, no one understands your job and the issues you face till you come to a meeting like this. Per which a hapless Silicon Graphics, Inc. (SGI) man is nearly lynched after publicly suggesting that system administrators wanting source code couldn't read it even if they had it--a lack of awareness typical of what site reps may face in the Greater Outlands, but do not expect to face at CUG. CUG is something else: a place to grind grits and gnash nits, where users talk and vendors listen and vice versa, where everyone respects everyone else and a lot of work gets done. Of course the epochal SGI/Cray Research merger did occur, and everyone is waiting to see if SGI will behave in a ravenous and Hunlike manner toward Cray, and if CUG will go to the dogs--but the outlook looks good, at least in beta. SGI chief executive officer Ed McCracken deems CUG important enough to personally attend. He addresses the members of CUG in a plenary session, where he is attentive to the organization and emphatically optimistic about the SGI/Cray future. His remarks show not only confidence but humor and even a degree of humility--which, if it is PR, is good PR. "I got no respect till I put 'CRAY' on my business cards," he quips. "People used to treat me like a salesman; now they treat me like a colleague." SGI's responsiveness to their new stepchild CUG is also embodied by Mark Goldman, who flies in heroically from SGI on a moment's notice to fill a program slot that has just opened. Goldman, manager of the Accelerated Strategic Computing Initiative (ASCI) project, tells the inside tale of how the newly merged SGI/Cray staff, eyeing each other with sweaty palms and trepidation, did a mind meld to win the Department of Energy's award for building a teraflops computer. (The ASCI concept: To stop bombs from blowing to bits in nuclear tests by turning bombs into bits inside computer models.) Meanwhile, Bo Ewald, president of Cray Research, Inc. (CRI) and senior vice president of SGI, is a strong presence. He attends CUG planning meetings and addresses two plenaries, talking straightforwardly about challenges and opportunities for Cray. He also attends the CUG receptions, where he mingles with delegates and welcomes first-time attendees with a graciousness and respect toward users that is touching. If it is PR, it is excellent PR. Vice president of CRI Supercomputing Irene Qualters steps up to the board and gives a presentation on the supercomputing roadmap. But first she tells us how she was riding her bike the day before and thinking about Seymour Cray's death, the purpose of life, and What It All Means. We feel like crying in our beer, then changing the world. Cray Research also missions a cadre of hardware and software and service people who give talks, offer help, answer questions, dispense advice, and tell how they were stung by bees on the golf course. Outside in downtown Charlotte the air is sweet and humid. Historical plaques line the walks; art sculptures and commemorative statues define every corner. Fountains are omnipresent--enormous cascading waterfalls, pools with ripples and dancing waves. At sunset, gleaming glass buildings reflect a fiery sky. It's good to get out once and a while. You have to sometimes, at least to escape the hotel coffee for a cappuccino next door. CUG is great. It gives you these weird opportunities. Like, everyone has seen Los Angeles, but who's ever been to Charlotte? Back in the business meeting, people wonder if, with the SGI/Cray merger, we should change the name of the user group; "SiliCUG" is suggested. Debates rage over bylaws and amendments; new officers take charge. The technical program continues. Presenters describe express-link options, source-synchronous links, and modular configurations. The audience asks about memory-read latencies and bisection bandwidths. Memory, bandwidth, aliens--it's been real. On the last day, David Robertson, local arrangements chair for the next CUG, invites us all to the San Jose meeting in May 1997. He promises that it will be an earthshaking event. "Silicon Valley is the epicenter of high technology, and this CUG will feature the latest upheavals in supercomputing. Leave your typhoons, tornadoes, and hurricanes behind and come to California for 'Seismic Supercomputing'! "Don't let the theme scare you away," he adds. "There hasn't been an earthquake in months."
"I said metarouters, not gigarings. Ha! Ha!"


We feel like crying in our beer, then changing the world.