A couple of recent announcements out of IBM once again pose questions about not only the future scope of parallel computing but also the timescale encompassed by that word `future’. 

The multicore Cell processor has so far made its name well away from corporate computing, being used in graphics-based applications in systems like Sony’s Playstation 3. A new version is just about to appear, however, and this will feature the addition of much stronger floating point capabilities. This is just what is needed to make it suitable for applications in areas such as large, complex financial calculations – things like risk management which, as our current economic situation seems to demonstrate well, is still largely at the Ouija Board/pint of brandy/take a guess level of incisiveness. 

As has been found before, where the financial community leads the rest of business starts to follow in the fullness of time. If the new Cell processor shows itself capable of giving financial calculations some real clout there could be growing pressure from business users to exploit its capabilities elsewhere. 

This could be supported by the appearance of a new idea from the company’s labs. We have all got used to hearing about High Performance Computing (HPC), the world of scientific and academic supercomputers where parallelism is a core part of the natural order. There has been a good deal of talk over the last few years about HPC making it down into the grubbier world of business and commerce, but there are some problems with this notion, mainly concerned with the fundamentals of the software architecture being geared to high level computational tasks. 

So engineers at IBM’s labs have come up with a different approach, High Throughput Computing (HTC), which is seen as being far better architected for the needs of business users. Though still a long way from being an official product, HTC is being lined up to create a software platform that can make parallel supercomputers like IBM’s Blue Gene architecture far more amenable to the commercial world. 

IBM is working with the Condor Project Team at the University of Wisconsin-Madison and a joint development effort is now in place to create the HTC environment for Blue Gene systems. One of the keys to Condor is its ability to match resource requests made by incoming jobs with resource offers from what is available in the system. It can even exploit unused desktop resources, and can work with either serial or parallel jobs. 

Such an environment could even help get round one of the fears concerning the coming of parallel computing in business – that while it is probably inevitable and will quite possibly prove to be a `good thing’ in the long term, the pain of getting from here to there could be severe and even damaging for some.  

If HTC can work effectively and contentedly with serial and parallel code, it could be the basis for a long-term bridge between here and there. If it can then work with Cell-based servers, which run Linux, the basis for a real stepping stone between serial/legacy architectures and parallelism maybe just around the corner.