Power Management, from Server to Cloud

With the 1st quarter of 2010 upon us, Intel is very focused on the launch of their latest Xeon processor.  This newest large scale design, codenamed Nehalem EX, will increase compute horsepower, address huge RAM memory allocations, and include all sorts of advanced RAS features for high performance computing in the mission critical space.  Yet amidst all of this hoopla, we might forget that data center managers are still grappling with the fact that they are running out of power and cooling capacity.  The need to balance performance with power consumption continues to be a balancing act in data centers across the globe.

Luckily, Node Manager and Data Center Manager continue to be supported with this latest Intel processor and you are still able to set very sophisticated policies to control power use within a system, rack, row, or even the entire datacenter.  The main focus of this technology continues to be allowing higher densities based on capping power usage within the rack.  When we are capable of regulating power use, we can design data centers more efficiently.  We reduce over-provisioning and limit stranded power. Data center managers benefit from the following use cases:

  • Energy cost rebates from provider based on the ability of a customer to not exceed a specific power consumption level.

  • Reducing datacenter hot spots through policies that are triggered based on thermal sensors.  The reduction of processing frequencies reduces power consumption and reduces thermal output.

  • Increasing server density in a collocated environment.  The ability to cap power gives greater confidence of maintaining a per rack power cap where a customer is billed by the rack for compute resources and required to stay below a certain power maximum.

Recently I have been working with a cloud computing provider.  Based off work Intel has done around power capping using Node Manager and real internet workloads, we have seen compelling data that demonstrates power savings while still maintaining performance SLA’s around response time, latency, and query success rates that ensure a continued quality end-user experience.  Also, based off customer generated costs per watt calculations, savings from power capping was estimated as high as $100 USD dollars per server per year.  Take those savings and multiply them by the hundreds or thousands of servers in these data centers and the savings become very exciting. Has anyone created their own cost saving estimates for the reduction in cooling, floor space, based on increased rack densities, and reductions in stranded power, when server power use is decreased 1w, 5w, or 10+ watts? . . . Something to think about in our spare time while we wait for the launch of Nehalem