There was an unaddressed question that danced around the edges of the Intel Software Conference held earlier this week in Salzburg: where is the potential for parallel technology to dramatically change the way users set about computing, particularly in the datacentre and the wider Exostructure? No real answer emerged, but given that the target audience was the European software reseller community, it was perhaps inevitable that such thoughts were considered too far into the future.
In practice, the event demonstrated an important watershed in the development of parallel computing beyond the esoteric realms of High Performance Computing and out into the mainstream. There the realities of 'what is' still rank higher than the potentials of 'what might be', and require developers to be still thinking in terms of the adaptation and development of existing applications that can better exploit the already available multicore processors, rather than dreaming up new services.
While this can be seen as simply thinking 'within the box' it does map onto the current, if not eternal, verities in that every server now being sold has a quad core processor and that the majority of servers in daily operation are at least dual core systems. In addition, though there is a strong presence of Linux and its associated applications in datacentres these days it is impossible to avoid Microsoft's position in that market, and the consequent need for Microsoft-based application developers to find ways of exploiting their existing skills and tools.
This is why one of the featured technologies at the conference was Intel's upcoming Parallel Studio, which is aimed at bringing parallel programming to client developers on Windows and maximising the productivity parallelizing C++ applications using Visual Studio on Windows. This is an onward development and packaging together of a number of development tools already available for the Linux marketplace, and is due to appear on May 26th. A more detailed review of this development will follow in ITA shortly.
For the short term both Microsoft and Intel are targeting developers working on the adaptation of the status quo. This is certainly understandable, but arguably lacking in an expression of a vision for even the near term—a point that was quietly noted by a few channel-based delegates at the conference. The applications bias, including the re-engineering of existing applications to exploit parallelism, while important, meant that discussion on how parallelism can be exploited to develop new ways of approaching the management and manipulation of data was missed.
There was little about parallelism and service development from a higher strategic level which was a shame, particularly as parallelization is liable to be one of the important technologies behind the further development of resources out in the wider Exostructure. It may not be that long before developers' constitute a relatively small coterie of specialists, developing the tools and process components that will be used by business managers to assemble their own services within the resources available from a service aggregator.
There was, however, one potentially significant nod in the direction of the future from James Reinders, Intel's Chief Software Evangelist and Director of Software Development Products, in the form of the Larrabee graphics processor, due to appear late this year or early in 2010. This will be Intel's first productized step into the world of 'manycore' processors, with the company already publicly—if indirectly through released details of simulation data—having pointed to a maximum of 32, x86 cores being available in the device.
The potential for manycore specialist graphics processors, such as those from Nvidia, finding alternative roles in business-related data processing has already been widely noted, so the potential for the Larrabee processor to find a similar role is already being acknowledged.
Such potential does, however, beg the question as to whether Microsoft applications will be able to transition to such an environment. Reinders does not feel it likely that anyone will be attempting to use Larrabee for business data processing any time soon. But it is a fair bet that some will be experimenting as soon as they can acquire a few devices, while a few may be attempting simulations even now.