The race for leadership of parallel processing into the future is now well underway, and there is a chance that Sun Microsystems – not exactly a dark horse but not necessarily the first name to spring to mind as a main contender – may actually steal a march over the rest.  But this does then beg the question: how much will this directly affect enterprise users? 

It’s upcoming Rock Processor makes use of an old technology, transactional memory, but in a new way that should make developing applications and services a good deal easier than otherwise with any parallel software development technologies, such as multi-threading so far available. What is perhaps more important, however, is that Sun is also amongst a number of vendors funding, to the collective tune of $6m, Stanford University’s new Pervasive Parallelism Lab. The others include IBM, HP, AMD and Intel. 

It does not, however, include Microsoft, which has teamed with Intel to fund
Berkeley
University and the

University of
Illinois ($10m each) on similar parallel processing research. Microsoft and Intel are, of course, prime movers in multi-threading. 

Transactional memory can be implemented in software or hardware, but both can require significant levels of skill. Sun has come up with Hybrid Transaction Memory (HTM) which sets out to exploit the best capabilities of both, making development work easier than extremely difficult, and offering the chance for applications performance to improve without change as the Rock processor is developed further. 

And as the Stanford lab is headed by Kunle Olukotun, who helped design chips that are part of Sun’s UltraSPARCTx family, it might be possible to guess the general research direction the Lab may take. As IBM is also a chip maker in its own right, transactional memory could get some reasonable leverage behind it, which might just leave Microsoft ploughing a different furrow alone.

But back to that question: fundamentally, how much should an enterprise care about such things? Well, if one assumes that traditional processor and software architectures are, while by no means dead yet, widely diagnosed as terminal; and if one also assumes that what follows is parallel processing,  then every CIO and IT director needs to be considering what may be coming next, what its impact might be and what will be involved in implementing it. 

These are archetypal `big questions’ with no easy answers as yet. There might just be one, however.  

Parallel systems are going to be large – the larger the better in a way. They are also going to be expensive to buy and to run. But they are also going offer prodigious levels of performance, both in the shear grunt of throughput and in areas such as I/O management and capabilities.  

As we all know, the likes of Google, Amazon and Microsoft are building huge datacentres aimed at the nascent utility services market. They would likely find parallel systems strangely attractive, precisely for the above reasons. Then they would need to market the hell out of the service offerings.  

This could lead enterprises to seriously examine the possibility of switching to utility services simply because the cost and hassle of trying to make the parallel move alone will probably prove too expensive and/or traumatic. 

It could be that no-one need worry about making the move for it will only be the big utility players that have the half-dozen rocket scientists that ultimately need to know all this stuff.