Spend a day now and again suffering death-by-powerpoint. It's instructive. It shows up interesting examples of massaging numbers to say just what you want to say. The old saw "lies, damn lies, and statistics" is not a reflection on statistics. They don't lie. People do, however, and they use statistics to do so. In one day of a conference, you can pick up half a dozen examples. I'll omit names, to protect the guilty, but here are some curious ones.
What they don't tell
Virtualisation is a way to increase efficiency, right? Nobody disputes that. You get better utilisation out of consolidated hardware resources, and gain a whole lot of other benefits too, even if the technologies to do so are still pretty bleeding-edge, and still only advisable to large organisations that can realise large gains for a large investment at large risk.
How much efficiency? A number that comes up time and again is that the average server runs at four-odd percent utilisation. This seems shocking, and suggests that a ten- or 20-fold improvement is a reasonable expectation.
Yet what none of the presenters pointed out - but several audience members who actually run live virtualisation projects picked up - is that the main reason processor utilisation is so low is because the RAM that goes with it is fully utilised. Many applications require a lot of memory, but only occasionally require full use of the processor. Without being able to split RAM into a separate pool, such as we do with storage thanks to fast data transfer technologies such as fibre channel, it is pretty hard to utilise a processor better when all the RAM in the machine is already in use. This is a non-trivial problem to solve, and will probably require a technical solution that connects a memory pool of arbitrary size - shall we call it a "memory area network"? - to an arbitrary number of processors. That lack of flexibility is just one possible bottleneck in hardware systems that makes processor utilisation a bad indicator of overall infrastructure efficiency.
Grand differences
How about power saving? Every presenter seemed to agree, and those in the audience running big datacentres confirmed this, that power for hardware and cooling represents about 10% of total operating costs. Some analysts, Gartner included, believe this ratio could soon become 50%, as power prices rise and technology costs decline, but let's work with the 10% number for now.
Relying on touchy-feely, speculative green benefits, using impenetrable, inconsistent and frankly rather dubious statistics, is not going to get vendors very far
Ivo Vegter is a freelance journalist and columnist.
A fellow from a large telco reckons he could get power costs down to three or four percent of operational expenditure - as saving of 60% or so. One of the vendors who presented said 60% is a reasonable expectation, though in some cases he's seen only a 30% saving on energy.
Yet a cost model for two levels of virtualisation - a basic level at 5:1 consolidation, and a more advanced level at 50:1 - showed power costs at 0.8% and 0.3% of the total budget, respectively. Ditto cooling costs, so total power consumption accounts for as little as 0.6% of the overall price. This suggests a total energy saving of 94% (from 10% to 0.6%) is possible. That's a grand difference from 30%, is it not? Such inconsistency does not do much for the credibility of the business cases the vendors present. This might explain why bean counters are so skeptical of green arguments for server consolidation projects.
Green pretences
Another vendor, aware it wasn't addressing accountants, did rave about its green credentials. The company encourages staff to work from home, saving who knows how much energy, especially in transport to and from work.
It didn't occur to them to explain the energy use of your average home employee, however. For a start, the company simply shifts some energy costs, along with the cost of real-estate footprint, to staff. This takes it off the P&L where it sits in plain sight of auditors and green advocacy groups, and means it can be reported as a saving of so many millions.
But each "telecommuter" also needs a dedicated printer, instead of sharing one between a dozen employees. Each needs to heat or cool their own house, instead of sharing space in a well-insulated, high-density office building.
The company claims to have done research which shows, implausibly, that home users use half as much energy for office equipment than they would do in an office, and that the total saving, all told, runs to 50 or even 100 times. This stuff didn't make it into the presentation, however. Perhaps because it is at best counter-intuitive, and possibly entirely false. Doesn't it owe an audience an explanation of how it arrives at such fantastic numbers, instead of simply ignoring them? And when it reports so many millions saved - impressive though this sounds - doesn't it owe the audience a denominator? A total cost of which those millions are just a fraction, so they can gauge the real scale of the saving, and the real pay-back period for these investments in going green?
Real benefits
Perhaps it would have been better served by simply pointing out that two hours less in traffic every day, and fewer interruptions, make home users happier and more productive. That's an argument that might make sense to directors and shareholders. It establishes real benefits.
Relying on touchy-feely, speculative green benefits, using impenetrable, inconsistent and frankly rather dubious statistics, is not going to get vendors very far. On the other hand, one of them did point out that the market for "green" products would reach $688 billion by 2010. Pitching a presentation to grab a share of that steaming pie might make sense after all.
-----------------------------------------------------------------------------------------------------------------------
* Ivo Vegter is a freelance journalist and columnist, who blogs at http://ivo.co.za/. He's as green as green can be, working from home.
Share