It seems to me…
In the very early 1980s, NASA initially had planned to construct another proprietary computing system similar to the obsolete ILLIAC IV, which had been built under contract by Burroughs, but to incorporate the very latest available technology. Requests for bids for the new contract were requested and received from several computer manufacturers. General Electric Space Systems was contracted to review the submitted bids and select the winning proposal. Since I previously had been employed at Control Data Corporation which, prior to Seymour Cray leaving CDC to start his own company, was the only company producing “supercomputers”, it was thought I might have some expertise in the selection process.
All the bids received were considerably over initial budget projections and consequently rejected. It eventually was decided that rather than building another proprietary system, NASA would purchase three of the latest most powerful computers available on the open market and replace the slowest about every three years.
As part of a totally separate unrelated contract, I had been a member of a team working on a proprietary system proposal for a distributed image exploitation application. Images stored on videodisk carrousels were to be located and displayed on the appropriate analyst’s workstation within a specified amount of time.
We initially were unable to meet the time requirements until I recommended scrapping our computer-centric design and reorienting it around a new network system similar to one being developed at Xerox PARC just up the road from us where I had a number of friends. (The network speed at that time still was not sufficient so we actually proposed a hardware switch in the initial proposal.) While probably not the first project to use this type of architecture, I wasn’t familiar with any other at that time and called it a “server-oriented architecture” in the proposal.
After submitting our proposal to the customer and prior to the actual contract award, I was contracted through General Electric Space Systems to write a Systems Concept document for the NASA Ames Numerical Aerodynamic Simulator (NAS) for the large supercomputing system replacing the ILLIAC IV.
Though I proposed a design fairly similar to the one recommended in the previous proprietary proposal, the System Concepts proposal approved by NASA is another example of how things are not supposed to be done.
I was the project leader of a small group of about four people assigned to write the NAS Systems Concepts document. We had written two fairly detail proposals but each was rejected without any constructive feedback as to NASA’s objections concerning the recommended design. The value of the contract was estimated to be about $309 million and I was very concerned about the consequences if our recommendations were rejected for a third time.
The night following rejection of the second proposal, my wife Barb, and I were at the home of Gene Greer, who was the NAS contract manager, and his wife Sue for dinner. Around 9pm after dinner, and at least one bottle of wine, talk turned to the Systems Concepts document. While Sue cleared the table, we started going over my ideas for the design. Gene grabbed some paper and started writing as I talked; Barb was given illustrations to draw. Sue kept making coffee while we worked.
Around 2am, we thought we might have another version of the proposal. I’m not sure what our administrative assistant, Chris Santorelli, or her husband thought when we called and asked if she would mind coming into the office to type up the proposal but she nevertheless agreed and met us there. While she typed the proposal, Barb redid all of our illustrations. I proof-read as anything became available.
The final proposal was finished around 9am the next morning and submitted to NASA. All of us went home for a nap. By the time we returned after lunch, the design was approved.
The basic difference between the two rejected design proposals and what NASA finally approved was elimination of many of the design details – there were considerably fewer details to which anyone could object.
Though the design was approved and implemented, it did not proceed entirely without question. While NASA was willing to try the recommended architecture, other reviewers — especially one at Dataquest — definitely considered it a mistake. (A couple of years later, while sitting unrecognized in the audience at a supercomputing conference in Santa Clara, CA, the author of the most critical article publicly admitted he had been incorrect in his assessment and apologized for the way he had totally dismissed the architectural design.)
It was not until sometime later that the system architecture was referred to as “client-server” though the numerous Silicon Graphics and Sun workstations in the NASA proposal had been referred to as “clients”. I would be interested in hearing from anyone with an earlier actual full-scale implementation.
Several years later when thinking about the basic design, I decided it could have been substantially improved by incorporating several peer-to-peer elements. It would be interesting to know if it could have had any affect on the way large computing systems normally are implement today. Paradigms being what they are, I wonder…
That’s what I think, what about you?