Comments on: Power as Ultimate Limiter http://www.douglasgourlay.com/blog/2009/07/power-as-ultimate-limiter/ Data Centers, Virtualization, and Cloud Computing Sun, 02 May 2010 01:04:25 +0000 hourly 1 http://wordpress.org/?v=3.0 By: Douglas Gourlay http://www.douglasgourlay.com/blog/2009/07/power-as-ultimate-limiter/comment-page-1/#comment-34 Douglas Gourlay Wed, 15 Jul 2009 20:06:09 +0000 http://www.douglasgourlay.com/blog/?p=46#comment-34 One thing I did not put into the main post, but is almost laughable that it is not implemented in many platforms is the following: If the ports are administratively shut down, why do the chips initialize and raw power? I.e. put a port in down/down state administratively. Why should the PHy and MAC power up? Why not keep as many system components in a low power draw state and have a sublinear power curve based on true-demand rather than having the device run 'hot'? Now I know why this is- its because its EASIER on the software developer in teh engineering team. But I think we should all start demanding more efficient and effective power management - a variable rotation fan is great, but there is a lot more that can be done. Let's start by measuring the power/throughput but in an apples/apples way based on the featureset. A switch with tunneling and MPLS and huge routing tables can not be compared to a switch with just forwarding features. Similarly Netflow accounting takes a large amount of transistors on the silicon die to support it at rate. Evaluate your true requirements for implementation, then baseline the system choices you have. Power isn't the only cost factor, but it is a significant part of lifecycle cost on these IT assets. Evaluate total cost (power, ops, maintenance, etc), then risk, and then quality and make a decision... dg One thing I did not put into the main post, but is almost laughable that it is not implemented in many platforms is the following:

If the ports are administratively shut down, why do the chips initialize and raw power? I.e. put a port in down/down state administratively. Why should the PHy and MAC power up? Why not keep as many system components in a low power draw state and have a sublinear power curve based on true-demand rather than having the device run ‘hot’?

Now I know why this is- its because its EASIER on the software developer in teh engineering team. But I think we should all start demanding more efficient and effective power management – a variable rotation fan is great, but there is a lot more that can be done.

Let’s start by measuring the power/throughput but in an apples/apples way based on the featureset. A switch with tunneling and MPLS and huge routing tables can not be compared to a switch with just forwarding features. Similarly Netflow accounting takes a large amount of transistors on the silicon die to support it at rate.

Evaluate your true requirements for implementation, then baseline the system choices you have. Power isn’t the only cost factor, but it is a significant part of lifecycle cost on these IT assets. Evaluate total cost (power, ops, maintenance, etc), then risk, and then quality and make a decision…

dg

]]> By: Douglas Gourlay http://www.douglasgourlay.com/blog/2009/07/power-as-ultimate-limiter/comment-page-1/#comment-33 Douglas Gourlay Wed, 15 Jul 2009 19:54:44 +0000 http://www.douglasgourlay.com/blog/?p=46#comment-33 To a point @aneel was making loudly on Twitter today we need to be able to synchronize the moves of Virtual Machines with their associated storage (or abstract that move from the workload) while maintaining network policy, finding the closest availabel default gateway, ensuring service consistency for transport in-band services like SLB and Firewalls, and coordinate these moves with the availability of sustainable and cost-effective power generation at a facilities and geographical level. The further we can move a workload, statefully, the more choices we have about where that workload can most efficiently be processed. Arguably a good problem to have. dg To a point @aneel was making loudly on Twitter today we need to be able to synchronize the moves of Virtual Machines with their associated storage (or abstract that move from the workload) while maintaining network policy, finding the closest availabel default gateway, ensuring service consistency for transport in-band services like SLB and Firewalls, and coordinate these moves with the availability of sustainable and cost-effective power generation at a facilities and geographical level.

The further we can move a workload, statefully, the more choices we have about where that workload can most efficiently be processed. Arguably a good problem to have.

dg

]]> By: Stefan http://www.douglasgourlay.com/blog/2009/07/power-as-ultimate-limiter/comment-page-1/#comment-32 Stefan Wed, 15 Jul 2009 19:40:27 +0000 http://www.douglasgourlay.com/blog/?p=46#comment-32 Actually "cloud" may not be the answer, but the unfortunate victim of this [power] limitation - i.e. if I cannot "move" I/S/PaaS across Data Centers, when in need to re-balance resources, because I may have plenty of CPU or memory in the servers in the target DC, but not enough power to sustain the new load, then I am really in trouble. On the other hand, looking at this type of problem as an opportunity, though, maybe we could come up with facilities controlled management tools which could have as much an impact on resource (automatic?!? = cloud) reallocation decision as the internals of servers and networks have today ... Stefan Actually “cloud” may not be the answer, but the unfortunate victim of this [power] limitation – i.e. if I cannot “move” I/S/PaaS across Data Centers, when in need to re-balance resources, because I may have plenty of CPU or memory in the servers in the target DC, but not enough power to sustain the new load, then I am really in trouble.

On the other hand, looking at this type of problem as an opportunity, though, maybe we could come up with facilities controlled management tools which could have as much an impact on resource (automatic?!? = cloud) reallocation decision as the internals of servers and networks have today …

Stefan

]]> By: Douglas Gourlay http://www.douglasgourlay.com/blog/2009/07/power-as-ultimate-limiter/comment-page-1/#comment-31 Douglas Gourlay Wed, 15 Jul 2009 19:35:10 +0000 http://www.douglasgourlay.com/blog/?p=46#comment-31 James, I have seen some companies even in what we would think of as 'hot' climates like Las Vegas use outside air economizers. There is a pretty substantial savings IF the outside ambient is lower than inside a reasonable amount of time. Turns out Las Vegas is cooler than the 80.6F Omar references about 70% of the HOURS in a year (but maybe not the daily high). You have to couple this with software, and intelligently managed HVAC that can recirculate, or pull from outside depending on the temperature for that particular HVAC system and its particular air. It's easier to imagine if you think closed-loop cooling like a nuclear reactor) dg James,
I have seen some companies even in what we would think of as ‘hot’ climates like Las Vegas use outside air economizers. There is a pretty substantial savings IF the outside ambient is lower than inside a reasonable amount of time. Turns out Las Vegas is cooler than the 80.6F Omar references about 70% of the HOURS in a year (but maybe not the daily high). You have to couple this with software, and intelligently managed HVAC that can recirculate, or pull from outside depending on the temperature for that particular HVAC system and its particular air. It’s easier to imagine if you think closed-loop cooling like a nuclear reactor)

dg

]]> By: Omar Sultan http://www.douglasgourlay.com/blog/2009/07/power-as-ultimate-limiter/comment-page-1/#comment-30 Omar Sultan Wed, 15 Jul 2009 19:24:05 +0000 http://www.douglasgourlay.com/blog/?p=46#comment-30 Doug: One interesting approach I have recently seen is to simply run the data center at a higher temperature. The American Society for Heating, Refrigerating and Air-conditioning Engineers just raised their recommended upper limit for data centers from 77F to 80.6F. With every degree increase potentially shaving 4% off DC energy costs, it certainly seems like an enticing approach, but I think you would really need to have a handle on your DC cooling before trying this. Omar Sultan Cisco Doug:

One interesting approach I have recently seen is to simply run the data center at a higher temperature. The American Society for Heating, Refrigerating and Air-conditioning Engineers just raised their recommended upper limit for data centers from 77F to 80.6F. With every degree increase potentially shaving 4% off DC energy costs, it certainly seems like an enticing approach, but I think you would really need to have a handle on your DC cooling before trying this.

Omar Sultan
Cisco

]]> By: James A. Schultz http://www.douglasgourlay.com/blog/2009/07/power-as-ultimate-limiter/comment-page-1/#comment-29 James A. Schultz Wed, 15 Jul 2009 19:17:07 +0000 http://www.douglasgourlay.com/blog/?p=46#comment-29 Any insight on what the ratio is between power consumed for cooling, and power consumed by co-located equipment? I've heard of data centers cooling through the use of external air when the climate is appropriate, but I have no idea how much energy this really saves. Any insight on what the ratio is between power consumed for cooling, and power consumed by co-located equipment? I’ve heard of data centers cooling through the use of external air when the climate is appropriate, but I have no idea how much energy this really saves.

]]>