POWERLESS?
It's the latest craze that's sweeping the nation -- sitting in the dark, powerless.
From California, apparently trying to take the Sunshine State title away from Florida, where legislation was introduced to restrict homeowners from using anything except efficient fluorescent bulbs in their houses (really, no kidding), to federal subcommittees investigating practical restrictions on data center power utilization, the newest hot topic is how to use less electricity.
It's no small matter. Data center owners, long concerned about HVAC capacity, are now seeing situations where they are not only unable to climate-control their facilities -- but they also can't physically run enough power to the buildings to add more servers. Power bills have now outstripped other operational costs by orders of magnitude in some situations, wreaking havoc on operating budgets and incurring the ire of CFOs.
As always, the pressure is "do more with less".
This is yielding some odd behaviour, from the slightly misinformed to the downright odd.
On the slightly misinformed side, hypervisors (VMware, Xen, and Microsoft) are now being pushed beyond their rational limits.
Don't get me wrong, I am a huge fan of hypervisors -- and for data center owners who are still running full physical servers at 7-8% capacity, shame on you, why haven't you virtualized! Go, now, and save power by combining 5 of those servers into 5 virtual machines running on one physical server.
That said, there are the folks in those situations who try to run 20 virtual machines on one box. If one pauses and does the math, it's immediately evident that to make that work, the physical box has to be a larger, higher CPU, more power-hungry machine. So claiming 20-to-1 consolidation is misleading -- they're actually only reaching about a 4-to-1 power savings ratio... and significantly raising their risk of disaster if the physical machine fails (unless they're running two in tandem, at which point the power savings is cut by half!).
Add to this the downright desperate approaches, such as water-cooled machines. Shades of the 1980's and giant Cray systems with fishtank bubblers! Maybe it's me, but the idea of deliberately introducing water into a data center seems, well, odd. I know these systems can be very successful, but at the same time, well, I've also seen what happens when one (one) breaks.
So my question is whether we're overlooking the obvious: why not simply turn off machines that aren't being used, and turn them back on as needed?
While it may be somewhat heretical, consider that this is the premise that powers many a hybrid automobile now cruising our nation's highways: when they aren't moving, the engine (gas and electric) turns off.
Most server-class machines are equipped with lights-out power management modules -- and those that aren't are being plugged into IP power strips.
So why not turn them off and on?
At Scalent, we manage power state as well as server stack, network connectivity, and storage access. Several of our clients use Scalent to conserve power by turning off machines until they're needed (server failover, web site or email surge), then turning them on and "cabling" them in real-time to meet shifting demand and needs.
They're saving money, they didn't need to modify existing infrastructure... and it seems a lot safer than water, eh? In addition, our California customers can take advantage of our status as a certified vendor, eligible for PG&E's virtualization program rebate -- making Scalent virtually (sorry) free.
Give us a shout. We'll show you. Until then, don't forget to turn off the lights on your way out.
-- Kevin Epstein, VP Marketing, January 2007
It's the latest craze that's sweeping the nation -- sitting in the dark, powerless.
From California, apparently trying to take the Sunshine State title away from Florida, where legislation was introduced to restrict homeowners from using anything except efficient fluorescent bulbs in their houses (really, no kidding), to federal subcommittees investigating practical restrictions on data center power utilization, the newest hot topic is how to use less electricity.
It's no small matter. Data center owners, long concerned about HVAC capacity, are now seeing situations where they are not only unable to climate-control their facilities -- but they also can't physically run enough power to the buildings to add more servers. Power bills have now outstripped other operational costs by orders of magnitude in some situations, wreaking havoc on operating budgets and incurring the ire of CFOs.
As always, the pressure is "do more with less".
This is yielding some odd behaviour, from the slightly misinformed to the downright odd.
On the slightly misinformed side, hypervisors (VMware, Xen, and Microsoft) are now being pushed beyond their rational limits.
Don't get me wrong, I am a huge fan of hypervisors -- and for data center owners who are still running full physical servers at 7-8% capacity, shame on you, why haven't you virtualized! Go, now, and save power by combining 5 of those servers into 5 virtual machines running on one physical server.
That said, there are the folks in those situations who try to run 20 virtual machines on one box. If one pauses and does the math, it's immediately evident that to make that work, the physical box has to be a larger, higher CPU, more power-hungry machine. So claiming 20-to-1 consolidation is misleading -- they're actually only reaching about a 4-to-1 power savings ratio... and significantly raising their risk of disaster if the physical machine fails (unless they're running two in tandem, at which point the power savings is cut by half!).
Add to this the downright desperate approaches, such as water-cooled machines. Shades of the 1980's and giant Cray systems with fishtank bubblers! Maybe it's me, but the idea of deliberately introducing water into a data center seems, well, odd. I know these systems can be very successful, but at the same time, well, I've also seen what happens when one (one) breaks.
So my question is whether we're overlooking the obvious: why not simply turn off machines that aren't being used, and turn them back on as needed?
While it may be somewhat heretical, consider that this is the premise that powers many a hybrid automobile now cruising our nation's highways: when they aren't moving, the engine (gas and electric) turns off.
Most server-class machines are equipped with lights-out power management modules -- and those that aren't are being plugged into IP power strips.
So why not turn them off and on?
At Scalent, we manage power state as well as server stack, network connectivity, and storage access. Several of our clients use Scalent to conserve power by turning off machines until they're needed (server failover, web site or email surge), then turning them on and "cabling" them in real-time to meet shifting demand and needs.
They're saving money, they didn't need to modify existing infrastructure... and it seems a lot safer than water, eh? In addition, our California customers can take advantage of our status as a certified vendor, eligible for PG&E's virtualization program rebate -- making Scalent virtually (sorry) free.
Give us a shout. We'll show you. Until then, don't forget to turn off the lights on your way out.
-- Kevin Epstein, VP Marketing, January 2007


0 Comments:
Post a Comment
<< Home