Life-cycle assessment of software products?

Question

I'm a software engineer, and lately I've been thinking whether it's possible to delineate the environmental impact of a software product - say a web service, a website, an app for your phone, etc.

It takes resources to create it:

• Developers, who use the company facilities while working on the software.
• Equipment - laptops, desktops, monitors, whatever else

It takes resources to run it:

• Cloud-based software runs in a datacenter, and enjoys the economy of scale a datacenter provides
• More traditional software runs on dedicated servers you have to manage yourself.
• A more inefficient algorithm may require more CPU cycles than an equivalent, but more efficient, algorithm
• Depending on longevity, this software may be running in the datacenter, or on people's phones, for many years

It also takes resources to deprecate/decommission software. Taken together, I wonder if there's a way to apply Life-Cycle Assessment (LCA) methodology to software products akin to how it's done to more traditional "physical" products. So much of our modern life runs on software that I suspect it's not going to end up being very cheap, when taken on its global scale.

Is anyone aware of efforts (or even theoretical studies) of applying LCA to software?

Thanks!

Answer 1

Easy.

If the software product is used by few, the environmental impact comes from your development and testing machines. Use low-power machines, as few as them as possible. Don't over-invest into high-end build servers, although if there are many developers, a single shared powerful server may be justified. The power consumption of these computers can be simply determined by a simple power usage meter.

If the software product is used by many, the environmental impacts of their computers dominates. Make the software as efficient as possible. You should consider the amount of memory used, and the amount of disk space used, although the main environmental impact comes from the amount of power used by the processor. Don't spin-wait, use efficient sleeping techniques. Use CPU-efficient algorithms and data structures. A fully occupied CPU uses easily 10x the electricity of an idle CPU. Look at the thermal design power (TDP) ratings of processors and you'll get the idea: they range from 10 watts in low-end laptops to over 100 watts in powerful desktops.

Some software runs in clouds. In this case, you should be as CPU- and memory-efficient as possible, too, to minimize the resource use of the software. Thermal design power is again a useful tool for determining the power use of cloud software. Don't forget to multiply the power used by the datacenter power coefficient, which takes into account the amount of auxiliary electricity used for e.g. cooling.

I'm not aware of any formal studies, but this is not rocket science.

Answer 2

You can easily conduct an LCA for a software product. What is probably most important is your underlying database though as I suppose you do not want to get engaged in primary collection. You can use open source software like http://openlca.org/. There, are also basic tutorials available for free on the website. The basic principle of an LCA is always the same you just have to use other data input.