The shape of servers to come

Here’s an interesting thought with which to start the New Year – are some server hardware vendors losing the plot? The recent launch by San Jose-based Supermicro of its 1U Twin server, which contains two dual-processor servers in a single 1U chassis, seems to suggest they are.

This architectural approach seems neither fish nor fowl in terms of the way server systems are likely to develop – and need to develop – in the next few years. On the surface, having four processors – and up to 16 processor cores – in a 1U box seems rather clever, especially given the better heat dissipation and energy consumption derived from sharing a single power supply unit (PSU) between two servers.

This may be an ingenious piece of industrial design and technology utilisation, but it still invites the question of why anyone would want to buy it. For a start, the single, shared power supply is an obvious weak point for companies supporting high-reliability, high-performance mission-critical server applications.

The 1U Twin server must surely come head to head with blade servers and, because it lacks the rest of the shared infrastructure integral to blade racks, it will almost certainly be found wanting. For example, squeezing two servers into a box where one would normally reside doubles the cabling at the back of the rack, creating more wiring spaghetti and causing no end of connectivity problems for IT engineers.

I am assuming that an important target market for the 1U Twin is companies that want to provide performance headroom for applications that need to scale-up rather than scale-out.

There is probably sales potential here, but the opportunity will slowly diminish. Even software that does need to scale up performance – essentially big applications like ERP – is being re-engineered into grid-compatible, SOA-based systems that can be clustered and scaled-out.

The 1U Twin design is also missing an important point: that servers will be required to provide different functionality to support new types of software in the future. An interesting trend in the development of business applications involves games-related technologies being harnessed for business-focused social software, for instance.

The way IBM is developing its Secret Island project on Linden Labs’ social interaction service, Second Life, is perhaps the best example of what might be possible here. On-screen “avatars” of executives who attend global project meetings without travelling, or support staff that conduct visual interpretations of product operating instructions, are just a couple of possibilities with real business potential.

Needless to say, server design will have to radically change in order to manage and deliver the prodigious quantities of high-resolution graphics to hundreds of users, all in real time. And that means much more than simply putting two systems in one box.