Daniel Robinson

It would seem that the mainframe is enjoying a miniature renaissance at the moment. Sales of these big beasts of the computing world are said to be on the increase, and as IT Week recently reported, IBM has released a new model in its Z series, which is claimed to be the most powerful of its kind ever.

This development could be seen as part of a bigger picture, a pattern that has been emerging for some time now, of enterprise IT beginning to migrate key functions back into the datacentre and bringing them under closer control of the administrator. While PCs and local area networks distributed business computing out into the hands of individual users, it might seem to some observers that enterprise computing is now returning to its roots.

Thin clients and server-based computing started the ball rolling for task-based workers, while more recently blade workstations, virtual desktops and a move towards web-based applications have in their different ways placed more emphasis on the datacentre and moved some users away from the fat-client PC sitting in front of each user. But I don’t think that this means the end of the standalone PC just yet.

IBM’s pitch for its new z10 mainframe seems to be one of consolidation. A fully loaded z10 with 64 quad-core processor chips and room for terabytes of memory is almost certain to take up less physical space and consume less energy than an array of PC servers with a comparable amount of compute power. Why not replace them all with the one big box that will be more reliable and easier to manage?

The problem with this proposition is that mainframes are still very, very expensive, which is why many customers often end up leasing them from IBM rather than purchasing them outright. A mainframe is basically a monolithic block of computing power, and while IBM might be able to turn up or down the capabilities of the z10 depending on what the customer requires ­ and is willing to pay for ­ investing in one is still going to look like a daunting purchase to all but the largest organisations.

Building your datacentre out of a battery of smaller discrete servers, on the other hand, is messier and less elegant, but has the advantage that it can grow along with requirements. You can start small, and add more servers as your company grows, which is exactly what many businesses have done. The end result may not be as power-efficient as a single, integrated big iron system, but there is almost always a trade-off with any technology.

I’m not trying to suggest that companies shouldn’t consider consolidating some of their infrastructure by investing in these new-style mainframes, but it is a move that is unlikely to make economic sense until your company’s datacentre has already grown to some considerable size. And even then, the z10 is unlikely to completely replace every single server or other system in a company.

As ever, reality is more complicated than a simple scenario of big iron versus the PC. While the mainframe vendors scoffed at the primitive capabilities of the early PC models, yet had to sit and watch while they slowly infiltrated every large company on the planet, so the PC enthusiasts that wrote the mainframe’s obituary have found themselves eating their own words in turn.

Both architectures are now mature and the strengths and weaknesses of each are now fairly well understood. It seems likely to me that both will have their place in enterprise computing for a while yet to come.