X
Home & Office

IBM's new eX5 server design: can you cope?

So IBM, inventor of the PC, with Tuesday's launch of the eX5 range, has finally decided that it's moving its nominally PC-based servers one more step away from the standard x86 PC architecture. In doing so, it has put clear blue water between it and the rest of the x86 server market, a gap that was narrowing as other players gained share through economies of scale.
Written by Manek Dubash, Contributor

So IBM, inventor of the PC, with Tuesday's launch of the eX5 range, has finally decided that it's moving its nominally PC-based servers one more step away from the standard x86 PC architecture. In doing so, it has put clear blue water between it and the rest of the x86 server market, a gap that was narrowing as other players gained share through economies of scale.

So IBM is back, with innovation that it reckons will significantly boost performance and cut costs, consisting essentially of bigger iron and more centralisation -- familiar territory for Big Blue. Just as a side thought, IBM may, as it claims, have spent three years engineering the systems, but you have to wonder just how long this innovation has been sitting in the lab, waiting for the right launch window.

For datacentre managers, the new eX5 architecture's key feature means more memory for bigger virtualisation hosts. With eX5, instead of a limit of 16 DIMMs, you'll be able to add up to an additional 24 DIMMs - using 16GB DIMMs that's 640GB of RAM per server, which should provide a fair amount of scaling. It comes in the form of a sidecar, named Max5, that seems to fit under the server, and is allowed because the CPU and memory have been decoupled.

IBM reckons it allow you to run 82 percent more virtual servers for the same licensing cost. No longer will you have to add another server because there's no room in the host machine's RAM to add more virtual servers. At this point, as long as the CPU isn't maxed, it's a waste of money and resources to add more servers; if you do, ironically enough you're likely to lower CPU utilisation rather than raise it, which is one of the key drivers behind virtualising in the first place.

The new servers will cover all the company's x86 ground, and with initial products consisting of a four-processor rack-mount, a new blade, and an entry-level two-socket server.

I suspect it'll be the savings in licensing that will ultimately drive this technology. Consider that the world's most popular enterprise database charges on a per core basis and you'll understand that the incentive to run Oracle on the fewest but most powerful CPUs available is strong, and IBM's Max5 helps enable that to happen. And as IBM points out, if you're running a 1,000-seat Microsoft SQL Server 2008 licence, you can cut your licensing costs by in half with the eX5 two-socket server.

There's other technology in there too, notable what IBM calls Flexnode, the ability to cut your server in half at times, so that you can, for example, keep utilisation up by running infrastructure applications by day and larger batch jobs by night on the same system.

It's good to see IBM innovating in this space. Now all you need to do is figure out how you'll handle all the additional I/O those memory-hungry databases will generate...

Editorial standards