Research Papers


    An Approach to Calculating and Defining Overall Data Center Energy Efficiency including Compute, Network, Storage and Facilities

    ABSTRACT The definition of data center efficiency in terms of the facility elements is well established. The universally adopted metric, Power Usage Effectiveness (PUE)[1] has several categories. There are two important issues associated with PUE. 1. PUE is not intended to take into account the energy efficiency of the IT equipment within the data center. 2. There are emerging trends toward locating some power and cooling functions that are normally part of the facility within the IT equipment. This obscures the demarcation between the facility and IT and can skew PUE results. In this paper a suggested framework for defining and calculating IT energy efficiency in terms of compute, network and storage in conjunction with PUE is discussed. The combination of these metrics results in a new metric for the overall data center efficiency. The principle that underlies the approach taken is predicated on knowledge of the energy efficiency of a particular device (server, network node or storage) for the operating conditions that the particular device will experience. In this context, energy efficiency is considered up to the point, but not including, the point at which data state change occurs. It is acknowledged that currently the gathering of data required to achieve this goal is not straightforward particularly for network and storage components. However most of the data already exists for compute in the form of OEM calculators. INTRODUCTION A holistic understanding of data center efficiency should take account of: 1. Compute environment workload 2. Data flows in networks 3. Data density and I/O rate in storage 4. Data center facility loads To be consistent with the approach taken with the PUE definition, determination of data center energy efficiency should not take into account system performance issues. Consider for example the PUE of a data center such as the Uptime Institute Tier 3 type and that of a Tier 1 type data center[6]. Inevitably the higher availability system yields a less energy efficient PUE all other things being equal due to the additional components and their associated fixed losses.