Raw computational power is being brought to bear on every stage of genome research from the moment the biologists tip their goo into a sequencing machine. They're used to read out the sequences, to knit overlapping sequences together into long deciphered strands of DNA, to find matches and annotations for the sequences if they exist, to observe and measure gene behaviour, and to measure the impact of drugs on the genes and the proteins they manufacture. The computers vary in scale from tiny DNA chips to supercomputers. The first computer is introduced at the sequencing stage. The millions of randomly-sheared single DNA strands have already been tagged at one end, and had a fluorescent marker added to the other. The marker is a special version of the A, C, T or G chemicals. The whole mixture is then put into an electrophoretic gel, which draws the fragments through at different speeds according to their size. As they arrive at the laser, the scanned images are passed to a computer, which extracts the sequencing data from the noise. These computers range from embedded microprocessors, through Macintoshes to Unix workstations.
The next few tasks blur together: overlaps between the sequenced fragments are used to establish longer sequences. These then need to be matched against public and private databases, which double in size every six to eight months and, if the sequences are new, they're submitted to the genome database. Phase two overlaps as well, as the growing genome database is mined to discover genes; the behaviour of the genes is monitored to establish their function, which means the proteins need to be analysed as well; and, as new discoveries are made, the genome databases are annotated. Furthermore, all establishments, public and private, need conventional computing capabilities, including the ability to serve the information needs of their peers in the outside world.
Several companies participating in the international HGP, including the UK-based Sanger Centre, and their commercial rival, Celera, chose Compaq for their projects by virtue of the speed of their systems. Celera, for example, asked several vendors to run a benchmark narrowing the field down to the only two players that met the task - one of those being Compaq's AlphaServer, which completed the task in seven hours. The other system took 87 hours. You may remember that Compaq bought Digital (DEC) along with this, inheriting not only the company's good reputation in academia but also its Alpha RISC processor design. The Sanger Centre's 150 DNA sequencers deliver gigabytes of information daily into the computer system for storage and analysis. It expects this figure to hit 20Gb daily quite soon. Since Celera has twice as many sequencing machines, and therefore it's reasonable to assume that the raw data volume is doubled, the consequent processing is orders of magnitude greater than Sanger's.
The Sanger system comprises 250 Alpha systems, from workstations to 16 Departmental Servers (DS20s) running Tru64 Unix, plus a further 48 Pentium PCs. Six terabytes of Compaq StorageWorks and Storage Networks RAID array systems are backed up by near-line and archive tape storage. The idea was to create a system that could scale at the drop of a hat, run continuously and handle faults without crashing or interrupting the calculations. The system also needed to be able to distribute the application load while avoiding 'turf wars' caused by researchers trying to run their work on other people's machines. Platform Computing's Load Sharing Facility (LSF) gives the IT staff the ability to balance the load between the various computers. For example, it allows all 250 machines to be dedicated at night to some of the processes, which users normally leave running for days.
Like Sanger, Celera has a network of AlphaServer computers running Tru64 Unix and TruCluster software to manage over 80 terabytes of data. The company's final assembly of the human genome required a 16-processor Global Server (GS160) with 64Gb of shared memory. It has spent a fortune with Compaq, around $US40 million. Apart from the Global Server mentioned, it has a further seven Global Servers (GS140s), more than 150 Enterprise Servers (ES40s), 14 Departmental Servers (DS20s) and 80 terabytes of fibre channel storage. In all, Celera threw 700 interconnected Alpha processors to the problem of sequencing the human genome. It was able to crunch the numbers in just seven months with a combined performance of 1.3 trillion floating-point operations per second (teraflops).
The most powerful computer in the world is soon to start work at the Lawrence Livermore National Laboratory in America. Called ASCI White and made by IBM, this machine has already been run at a mind-boggling 12.3 teraflops. That's 1,000 times more powerful than Deep Blue, the machine that defeated world chess champion Gary Kasparov in May 1997. This will be used to simulate explosions of America's ageing nuclear warheads. One bonus is that the scientists will now be able to see what goes on right inside an explosion. Of course, they can't see enough - they never can - so IBM will deliver an even more powerful machine in 2004. Apparently, 100 teraflops should do the job, according to experts. In other words, about one order of magnitude more powerful than the computer that's just been delivered.