The environmental impact of running software

This audio was created using Microsoft Azure Speech Services

The Greenhouse Gas Protocol is creating an accounting and reporting standard for the emission of greenhouse gasses from running software. This may be the first step down the long path towards a more fine grained energy efficiency metric than PUE.

GHG Protocol Product Life Cycle Accounting and Reporting Standard ICT Sector Guidance, currently in Draft status, concerns calculation of greenhouse gas emission from a number of different sources, software being just one. In this blog post, I will concentrate on the emission from software, and – to be more specific – the emission from running software, as opposed to the emission created during the development of the software. The recommendation for this can be found in section 7B of the Greenhouse Gas Protocol.

The key to calculating the emission of greenhouse gasses from running a piece of software is to determine the power consumption of running the software. When doing this it is critical to understand, that the environmental impact of a piece of software is closely tied to the power consumption of the hardware executing the software, and the operating system of that hardware.

Running a piece of software on Windows on an x86 based server may result in a different power consumption from running the same piece of software on Linux on a RISC-based architecture.

Although such measurement may seem simple, the task of identifying the power consumption of one piece of software on a system is complicated by a number of challenges:

• Operating systems improve their efficiency over time by learning from the usage patterns. For instance, Windows 7 usually spends the first days after installation optimizing the power usage, and not until at least three days after installation, can a consistent power measurement be made.
• Virtualization causes several operating systems and software application to run on the same physical hardware. Software running in one instance, may affect the overall efficiency of the host, and thus the power consumption of software running in a different instance on the same host.
• Defining the typical use scenario of the software to apply during the power measuring can be difficult, as the software may experience different usage patterns over time.
• The impact of inter-software communication requires several measurements. For instance, software making a remote procedure call to a data base server, causes power consumption on the database server as well as on all network equipment on the path between the two servers.

Although the Greenhouse Gas Protocol addresses several of these issues, it is still a long way from proposing feasible answers to them. The fact that the questions are identified and starting to be debated is, however, very valuable, and the work going forward will definitely increase the understanding of the power consumption in data centers.

If you have any thoughts on how to address some of these challenges, or even if you see more challenges than mentioned here, please engage us using the comments section below.