Cisco Blogs

How we can all drive Data Center Power Efficiencies…

- July 18, 2007 - 7 Comments

Good day all. I was just in a series of customer and analyst meetings where I was asked a couple questions about how can a customer drive more efficient use of their data centers. One person adroitly asked, “What can they do today?”I thought this was a good topic to see what everyone else thinks. I wrote a few ideas, they are certainly up for debate and I don’t pretend they are the be-all end-all answer. But if anyone has a different opinion, add your ideas, see what others come up with and comment on those too. What can we all do better? Here’s a few thoughts I have…1) Run your Data Center at 220V-240V, not at 110V. Credit for this has to go to Andy Bechtolsheim at Sun Microsystems who mentioned it to TOm Edsall and I a few months ago. This could drive a pretty quick 10%+ efficiency gain and not require all new servers, storage, or networking equipment. Big win.2) Look for silos of equipment. Do you have 20 Load Balancers? Why? Could One or two do the job from a throughput perspective? If so then look at virtualizing these silos. This is especially easy in security and Application Networking where the silo model seems to be most prevalent.3) Direct Attached Storage rotates as fast as a disk drive in your SAN – but you generally use a lot less of it. In some cases it can be more effective to go with diskless servers and SAN-boot them. 4) Older power supplies. Sometimes on older equipment that may be 5 or 7 years old or sometimes older the power supplies were designed with state-of-the-art efficiencies then of 80% efficient conversion. Some may have decreased in effiiciency as they aged or accumulated efficiency reducing dust/dirt or other airflow blockage causing them to run warmer. Maybe replacing them with a newer 90% efficient P/W makes sense.What other ideas do you have?dg

Leave a comment

We'd love to hear from you! To earn points and badges for participating in the conversation, join Cisco Social Rewards. Your comment(s) will appear instantly on the live site. Spam, promotional and derogatory comments will be removed.

In an effort to keep conversations fresh, Cisco Blogs closes comments after 60 days. Please visit the Cisco Blogs hub page for the latest content.


  1. #1 on your list is one that our CTO Neil Rasmussen wrote about. I helped with the analysis and found that the biggest electrical efficiency gain in going to a higher voltage distribution architecture is due to the elimination of the distribution transformers. The paper can be found at this link:"

  2. Good feedback Victor, are there others? What are your thoughts on having the PDU being outisde of the building and using the heat for regeneration? I heard that idea discussed before.Also one other piece I heard chatted up is raising the temperature in the DC. From a semiconductor perspective it may not make a lot of sense (a warm chip draws more power than a cold chip and doesn't have as long a lifecycle due to floating current increase if I remember right). Is this a valid point?dg

  3. Platform consolidation and virtualization offer the benefit not only of reduced power consumption but also of space reclamation.What is Cisco doing to reduce the amount of power required by its chassis, reduce it's watt/port ratio, and/or improve the efficiency of its own power supplies?

  4. On top of trying to stick with 220-240v power, even just running hot and cold rows can really make a difference in the cooling required. I agree that virtualization is great for compacting down the server bulk, but it can also introduce some unneccesary overhead in my opinion. Depending on the application, say load balancing, I don't think it would be wise to combine the jobs of multiple load balancers onto virtualized machines. Keep in mind that if the load balancer ever has issues, all of the servers directed by it are going to reflect those issues.

  5. Morgan, I can appreciate your thoughts on fate-sharing for some services like load balancers and firewalls, etc. In some environments discrete equipment makes perfect sense. Why though would we hold a load balancer to a significantly differing standard than the actual application that is being executed on a server? especially if the application is monolithic and not active-active wheras the load balancer is stateful active-standby or active-active configured?I raise this because I saw a customer recently who had 3000 load balancers for the exact reason you mention above. Average utilization 5%. Another had a requirements for 1000 firewalls, 1000 load balancers, and some inordinate number of switches for fate-sharing purposes. By tiering the applications and assigning a logical distribution of asset against resource they were able to reduce this number dramatically and save 310kwh. Over a 3-4 year operating period for the IT implementation that's quite a bit of $$$.dg

  6. These are all great points and should make it clear that driving power efficiency into a data center operation is a culmination of multiple solutions"".Many people often jump straight to the component level but should realize that while looking at the components is important, this issue is really 80% process and 20% product. The majority of data center products from the large IT vendors have AC power supplies that are in the range of 90% efficient. It is how these products are specified, deployed, managed and decommissioned that is where the big gains can be realized.A good example of this is server and storage utilization. Some analyst figures suggest that utilization of servers in US enterprise data centers in the range of 20% and storage only slightly higher at 38%. However, data center managers have the ability to drive utilization up through virtualization today. A key challenge to taking advantage of the power benefits of virtualization has to do with aligning facilities to accommodate more dynamic power and cooling requirements. A simple way to address this challenge is to bring on facilities staff onto the IT payroll.Brining in facilities skill sets into the IT organization can have profound benefits in aligning facilities supply with IT demand which is ultimately where the biggest efficiency gains can be realized.In addition to transforming the organization to include facilities on the IT payroll there are several best practices that users can take advantage of today and to get started on driving efficiency into the data center operation. We detail these in the Cisco White Paper: Energy Efficient Data Center Solutions and Best Practices. A summary of the best practices:1) Perform Regular Utilization Audits2) Bring facilities onto the network3) In-source facilities planning into IT4) Establish a Change Advisory Board for Green5) Monitor power consumption and thermals via the network6) Develop power profiles for services and infrastructure7) Develop a green scoring framework for services and infrastructureThe good news is that if the costs of power are considered by IT, there is a strong ROI for driving power efficiencies along with benefits around risk and capacity.Robert AldrichCisco Energy Efficient Data Center Solutions"

  7. [...]I have never heard of this trick before! But whoever you wrote this post about is surely crying right now…[...]