There's a battle going on for the hearts and minds of datacentre managers, and it's been framed as a question of blades. In essence, it comes down to whether a high density datacentre is more cost-effective than a less dense one, with blade servers as the new weapons.
On the one hand, we have the likes of Intel's Michael K Patterson, who argues that compute capabilities in the datacentre are increasing at a much higher rate than the total cost of ownership. Therefore, he argues, the consequence is that the cost per unit of compute performance is reducing.
Patterson reckons that there's a belief that cooling a high density system is much more expensive that a lower density one built out of standard rack servers. Consequently, TCO reduces because you don't need as much a/c plant or energy to keep the boxes cool. That's the theory.
However, Patterson cites an Intel study in which the processor company compares the cost of building and running two theoretical datacentres, one built using blades. The blade server datacentre costs less, reckons Intel because you need less floor space. Surprisingly, it also costs less per server if they're tightly packed because the static pressure needed to drive cooling air though the racks is lower for densely packed servers. That's because you'd use a high raised floor and more perforated tiles.
What Patterson doesn't factor in is an issue that both IBM and HP cite as an important saving when deploying blades: the reduced cabling plant. With maybe a dozen cables issuing from the rear of a blade chassis that contains well over a dozen servers, you've saved a lot not just in cabling itself but also because of the reduced cabling management load.
Additionally, blades are better connected than they were, with increasing numbers of blade-specific switches coming onto the market, as well as technologies such as HP's Virtual Connect, which allows MAC addresses to be virtualised. This gives you more flexibility, especially when managing virtual machines. More ports and more scope for virtualisation, together with improved cooling, could help sway the fiscal discussion in the direction of blade servers.
It's not quite as simple as that of course. You need to deliver more power per rack if it's stuffed with blades and this can be a problem: is there enough power getting into the room? Can you even get any more power from your utility company? These are crucial questions that demand an answer.
What's more, not all datacentres start with a clean slate. Often, you're stuck with the space and equipment you have, and installing racks of blades in one area of an already occupied server room can cause problems.
A rack of blade can draw up to 20kW, while the equivalent space filled with 1U servers might only pull one-third of that. So the blades are likely to unbalance your cooling arrangements to the point where you have to consider adding cooling especially for the blade-filled racks. Products such as APC's InfraStruXure can help in such situations, as they deliver targeted cooling to the point where it's required, rather than a more open space.
Of course you can go some way towards improving cooling, according to researchers Focus Consulting by ensuring that the underfloor area is clear -- just cut that cabling bloat -- and ensuring that cooling vents and blanking panels are appropriately sited. Here you probably need to bring on board specialists with the right expertise, experience and probably CFD software to calculate whether existing cooling systems and other infrastructure can cope with the added load that a blade-based installation will produce.
Another issue that blade vendors aren't keen to talk about, but which they will admit could be a problem, is the blade server vendor lock-in. The vendors' business model is akin to that of printer cartridges: the chassis are cheap -- probably at or below cost -- but their contents are not. What's more, you're locked into the chassis vendor's product set, which many enterprises could find a problem. One blade maker's product manager told me that many businesses will double-source its blade chassis purchases, buying one from HP and another from IBM, just to avoid this problem -- although this of course brings its own set of issues.
All that said, however, the direction in which most enterprises are moving is toward blades. At around 30 percent annual growth, it's the fastest growing part of the server market so blade technology clearly makes sense for a lot of organisations.
And since all those companies will have done their sums before committing, you might conclude that blade servers do offer TCO savings, as well as the more obvious benefits such as higher density and floor space saving.
However, right now, it's mainly large enterprises who are in a position to take advantage of blade servers. Assuming that small to medium-sized businesses are not likely to spend large amounts of money on the kinds of IT infrastructure that blade servers demand, it be interesting to see the reception afforded IBM's six-slot BladeCenter S chassis aimed at SMBs. It's due to start shipping later this year.
And it wouldn't be much of a guess to presume that HP, IBM's arch-rival and currently blade server market leader, is planning something similar.
Blade servers - will you be able to say no to what appears to be an inexorable trend?