Cooling future cost inflation

Waxman-Markey is great news for Sun - because doubling American energy costs within two years will first drive people to data center consolidation and then, finally, to facing the reality that 80 Watts PCs on user desks are heat producing power hogs that can be replaced with 4 watt Sun Rays without any change to the server software they've already got in place.

Servers using Intel's latest processors and the motherboards that go with them are among the most efficient, in terms of work done per unit of power consumed at full utilization, yet produced by the x86 industry.

Unfortunately a lot of that efficiency results from scale, and thus from workload consolidation and high system utilization - to quote Sun's bmseer:

SPECpower_ssj results shows that servers (even those with the industry's best power-management) running at low-utilization levels use many times more watts per unit-of-work than systems running at higher utilization levels. Datacenters can realize the biggest energy savings by running fewer servers at higher utilization levels (50% utilization or above).

Sun's results on the 8GB (or 0.5GB/core) configuration show that running at 10% utilization requires 4.4 times more power per unit of work than running at 50% utilization.

4.4 times = (581 performance-to-power @ 50% utilization /133 performance-to-power @ 10% utilization)

Most SPECpower_ssj2008 are published on small-memory configurations that are much smaller than typical customer deployments. Sun is the only vendor to publish multiple results to clearly show effect of memory configuration.

A more normal-sized memory configuration of 32GB (or 2GB/core) uses 30% more watts than a tiny 8GB (or 0.5GB/core) configuration at 100% load. At active-idle the wattage difference is also 30%. Some competitors use additional configuration differences such as non-redundant fans, non-redundant power supplies, and single slow disk to further reduce the wattages and significantly improve SPECpower_ssj scores.

He's complaining about competitors misleading customers, but the more important thing here is the extent to which utilization rates affect power efficiency - that 32GB, 16 core, Netra 4250 burns 225 watts doing nothing but making heat you also have to pay to remove -and that minimum only goes up by about 71 watts (31%) as you add workload until you max out the machine.

Now given the weathermen's commitment to more than doubling American power costs over the next eighteen months, the obvious conclusions you should be reaching from these numbers include:

  1. this may be the best argument yet for using OS ghosting in system consolidation and virtualization because, particularly for the Windows enterprise, this approach offers the lowest risk route to high utilization and thus lower power input requirements per unit of work done; and,

  2. it's also the best argument yet, particularly for the Windows enterprise, for replacing as many small servers as possible with virtualized instances running on much larger data center servers - because the limitations of Wintel software mean this is the only way to take advantage of the power savings available from larger, multi-core, gear.

Since these directions are consistent with what a large part of the industry wants to do for other reasons (mostly having to do with the consolidation of a different kind of power) the new energy taxes will, I expect, simply but radically accellerate an existing trend.

There are, however, several big negatives to doing this:

  • First, services to users usually decline with consolidation - and usage flexibility, often the key to corporate operating success, usually flatlines just as soon as the data center gains enough control to become the tail wagging the corporate dog;

    Saving power dollars at the expense of business flexibility may, in other words, illustrate the adage about being penny wise and pound foolish.

  • Second, data center consolidation within the client-server architecture does not address processor duplication - taking workload away from the desktop and giving it to servers doesn't significantly reduce desktop operating costs. Consolidated client-server is therefore architecturally similar to the mainframe systems people hated in the 70s - just without the simplicity and reliability that came with IBM's terminals.

    In other words, it's not wrong to reduce server room power use, but the big opportunities for savings aren't in the data center, they're on the desktops - in fact if data center management measured system wide CPU utilization they'd quickly realize that every successful move to increase data center utilization through consolidatation actually reduces system wide processor utilization.

  • Third, the long term IT staffing change is minimal in terms of head count but devastating in terms of skills availability and business continuity risk.

    Two things seem to happen:

    1. First, companies undertaking consolidation usually seek savings in help desk and related services but don't realize them in the longer term as the absence of desktop change combines with increased communications complexity and user complaints to quickly drive those costs back up; and,

    2. Second, the increased IT focus on data center operations means that fewer people run more of the business critical systems, and, more importantly, that those survivors are increasingly cut off from the business.
  • Thus what happens is that consolidation reduces the rate of externally driven systems change in organizational IT services without reducing the external presure for change. As a result business management works to find ways around real or perceived IT roadblocks and eventually, of course, IT management has to respond to the presure by doing something -but then discovers that there's nobody left on the bench to do it, and, worse, that the people kept on for their fine work in the data center don't speak user.

So is there a better idea? Something that responds effectively to exploding power costs while not incurring these negatives?

There is: and it's the opposite of what most people will be doing: instead of consolidating everything to the data center, get rid of the data center. Move your processors and IT staff into user spaces, replace as many 80+ watt desktop computers with 4 watt desktop displays as you can, and use Unix, not ghosting, to run as many applications as you need on each box.

The downside to this approach is that it's hard to carry out and unpopular with your peers, but the upside bottom line is that there are huge advantages for your employer in doing it - starting, not with the cost savings you'll achieve, but with having IT actually work with and serve the people who make the money, not the people who spend it.