In my last post, I wrote about the basics of understanding energy efficiency metrics, and why they are critical for IT organizations in determining datacenter inefficiencies that can lead to improvements and cost savings. Today I’d like to share some tips that can help set datacenters on a greener path.
Remember from my last post that a PUE of 1 means that all incoming power to the datacenter is being used for IT load. Let’s assume that we have done everything we can and achieved a PUE of 1.2 for a datacenter. What does this mean? That means nearly 83.3% of the incoming energy is utilized for IT load, which is pretty good, and a major improvement over the average. Now the question is how do I use this energy most efficiently? The obvious first answer is to implement virtualization. Yes, that is one solution. Then, implement automated provisioning and de-provisioning of virtual servers across multiple geographical locations based on the type of energy source and cost of energy. Implement policy-based server management tools based on energy data based policies and use of Wake-On-LAN and other PoE (Power over Ethernet) technologies.
If we take this to the next step, we can reduce the server power and associated heat generated by taking advantage of dynamic frequency scaling. This is also called “CPU throttling.” Dynamic power is directly proportional to the switching frequency and the input voltage, so if we reduce the frequency and/or the voltage, then the power will be reduced as well.
By following these steps and procedures, we will not only be able to reduce the overall energy consumption and increase energy efficiency at different levels, we will also be able to minimize the environmental impact directly. For every 1kWh of energy spent, nearly 0.6kgs of CO2e is emitted (according to UK standards). So, by minimizing the energy consumption, we will be able to reduce the footprint of the datacenter and related carbon emissions.
As many datacenters host mostly mission critical applications and services, availability is a key factor. Let’s talk about change management for a second. The primary objective of a change management process is to make sure there is minimal service interruption when changes are made in an IT environment. How many IT organizations consider energy data points in their change management process today? Not many. So, if a new server or router is provisioned, and if the rack is already reaching its maximum on the input power, we have a huge problem. Even with the approved change request, there is a danger of bringing the entire rack down. You can enhance availability by adding energy data points into the change management system, as well as other ITIL processes.
Let me share another example that really happened in our datacenter here at CA. Our service desk received complaints from users about intermittent Internet connectivity. When handling this issue, the network team suspected that the power input for the high available routers that act as the default gateway might be the issue. A report of the power utilization for the breakers associated with those routers showed that - yes indeed - the power input was having major fluctuations. This started on a particular late night on both the routers. Referring back to the change requests, they were able to confirm that at that time, there had been a firmware upgrade on these routers, which caused this power consumption pattern. This fact proves that adding energy data points to incident management or problem management can not only improve your MTTR (Mean Time To Recovery), but also can help keep your operations well within any service level agreement terms.
Therefore, maximizing energy efficiency in datacenters can do great things from both the business and environmental perspectives. In the case of a datacenter with a $1M energy spend, if we can improve the PUE from 3 to 1.2, we can achieve nearly $500K in cost savings. This new found funding can now be added to our operational budget or could be used to help save a few jobs in this tough market.