The IBM supercomputer that beat Kasparov. What was that system?
IBM Scalable POWERparallel platform presented in 1993 under the name SP1 as a supercomputer solution for scientific and technical calculations. The first versions of the platform were assembled from RS/6000 Model 370 servers. The development proved popular, and corporations and government organizations from the United States began to purchase it. For example, the Argonne National Laboratory ordered a system of 128 nodes. By the end of the year, 72 such platforms had already been sold.
A year later, IBM introduced an improved system, SP2. It used upgraded servers and a faster network connection called Trailblazer. SP2 proved even more popular, selling 352 platforms.
In 1996, SP2 was renamed SP. The platform began to be assembled from RS/6000 servers. At the same time, nodes began to be equipped with several processors and SMP multiprocessors. In 1996, IBM sold 3,770 RS/6000 SP systems to customers around the world. The Deep Blue series of chess computers were also developed on the basis of this platform.
Okay, what does Deep Blue II have to do with this?
Deep Blue has forever entered the history of chess: their opponent was one of the best chess players in the world, Garry Kasparov. In 1996, Kasparov beat the computer with a score of 4:2, and in 1997, he lost – the score was 3½:2½ in favor of Deep Blue.
Deep Blue II was an RS/6000 SP supercomputer. The system included two racks with 30 nodes – RS/6000 workstations.
And here is what each of them was based on:
P2SC processor: single-chip design of the chip with POWER2 architecture;
two expansion boards with eight dedicated chess processors on each board for the MCA bus.
The supercomputer had a total of 480 chess processors and 30 P2SC chips. Two servers were based on 135 MHz P2SC chips, and the remaining 28 were based on 120 MHz P2SC chips. Each node had 1 GB of RAM and 4 GB of disk memory. The chess processors were manufactured using 0.6-micron CMOS technology, and each had about 1.5 million transistors. The system was controlled by the AIX 4.2 OS.
Deep Blue consisted structurally of two parts:
software part implemented on a universal computer and performing calculations for the first few depth moves;
hardware based on fast specialized chess microprocessors. During further calculations they additionally accelerated and deepened the search.
This solution allowed the full use of the capabilities of chess processors. To make the system even more productive and flexible, the developers added the ability to connect external FPGAs, which adjusted the hardware search and evaluation. But by the beginning of the match, nothing was connected: there was simply not enough time. However, this did not prevent the supercomputer from defeating one of the best chess players in the history of mankind.
Not just for playing chess
Of course, the RS/6000 SP supercomputers were used for much more than just chess. They were used in areas such as computational chemistry, accident analysis, electronic circuit design, seismic analysis, reservoir modeling, decision support, data analysis, and online transaction processing. The system worked in such a diverse range of applications due to its flexible architecture, high performance, and scalability.
RS/6000 SP could be expanded to 512 nodes – Kasparov played with a system of 32 servers. Nodes are installed in racks, up to 16 servers in each. The SP platform could linearly scale disks along with processors and memory, and this made it possible to work with terabytes of file space. Also, parts of the system could be profiled, that is, “sharpened” to solve different problems. So, several nodes, for example, acted as Lotus Notes servers, and all the rest processed parallel database.
Each server runs the AIX operating system and other standard RS/6000 software. Software it was necessary installed on each server separately. To facilitate this process, IBM provided a set of software called Parallel Systems Support Programs (PSSP). It simplified the installation, configuration and administration of the system.
It is clear that centralized management of such a large-scale system is not an easy task. To solve it, IBM used one graphic console. It displayed the status of the software and hardware, information about the tasks being performed, and active users.
Using a single console, the system administrator could easily change user rights, keep track of tasks performed, manage printing, and conduct general monitoring.
What happened after Kasparov
The system continued to evolve after an improved version was created in 1997. In early 1999, IBM introduced an even more powerful version of the RS/6000 SP. This time, it allowed for the unification of not 512, but thousands of servers. The switch developed by IBM could handle more than 32,000 nodes.
The largest system was installed in the late 1990s by Lawrence Livermore National Laboratory, a national laboratory of the U.S. Department of Energy in Livermore, California. The cluster consisted of 1,463 nodes. No one has ever installed a system on an even larger scale: the cost of servicing such clusters would have been prohibitive.
RS/6000 SP of 30 knots worked as the main web server for the 1996 Summer Olympics in Atlanta and the 1998 Winter Olympics in Nagano.
The RS/6000 SP platform was supplied to IBM customers until 2001. Then there was a rebranding, and the supercomputer line continued under the name eServer p690. But that, as they say, is a completely different story.
That's all for today. And if you know any interesting facts about the history of this and other supercomputers, write in the comments.
