Clusters' first stand
It's easy to spot trends in IT -- if Dell and HP are doing it, it's a trend. The announcement that both companies are shipping new clustering servers is significant, even without the additional news that other players in the area are getting significant amounts of investment. Clustering -- the use of multiple computers and storage devices to create what seems to be a single system -- is coming out of the engineering lab and into the enterprise. You no longer need a lab coat to use it.
This is significant for many reasons. It is another piece of hard evidence that the future of computing is distributed -- the core model of one processor running a discrete set of applications is becoming outdated, and fast. It also shows that one of the great motors driving computer performance -- the cross-fertilisation of supercomputer design techniques with desktop technology -- remains a powerful force for innovation across the board.
Most excitingly, clustering's march into the enterprise highlights some of open source's greatest strengths and Microsoft's greatest weaknesses. Windows starts off with technical and strategic drawbacks that Linux has been swift to exploit: clustering gets much of its reliability and flexibility advantages from the idea that you can throw more processors onto the fire as you need more steam. With open source, that's all you need -- more chips. With Windows, you also need to manage and pay for licences: a pointless and expensive overhead that only acts to the user's disadvantage.
Technically, Windows still suffers from its origins as a single-user system coupled to complex and proprietary user and applications interfaces -- basic design ideas that work against the idea of having tens, hundreds or thousands of lightly coupled copies of operating system running in collusion to help a heterogeneous mass of tasks. Linux and the other scions of Unix have such concepts at their heart, and benefit from years of concentrated shared research into the practicalities.
Microsoft is very aware of this, and is working hard -- we are led to believe -- on its High Performance Computing (HPC) Windows variant. There are some massive clusters already running Windows: it can be done. Yet there is little or no commercial interest at enterprise level, where Linux is the incumbent with overwhelming presence and massive inertia. To compete, Windows will have to become more like Linux in the two places it matters most: freedom from onerous licensing and openness of interface.
Clustering has clear benefits for cost-effective and reliable enterprise computing, thanks to its openness and flexibility We trust that these ideas will rub off on everyone who decides to join in.