I focused my calculations on three areas: natural gas usage, employee commutes, and server electricity usage. It was not necessary to include office electricity usage because that comes from clean sources such as wind power.

The natural gas calculation was by far the easiest. Because we had historical usage data in therms, I could calculate metric tons of carbon released (the standard measurement when purchasing carbon offsets) by simply multiplying the number of therms by 0.0053. I then totalled all months in 2019 to get rough annual carbon production. Easy.

Figuring out the impact of employee commutes was somewhat more challenging, but still pretty straightforward. The factors I took into account included distance travelled, vehicle fuel efficiency, and the number of days commuted per year, allowing for holidays, vacation days, and work from home days. Data for these factors yields the number of gallons of gas burned annually for commutes to that office, and that in turn can be used to calculate pounds of carbon burned by simply multiplying gallons burned by 20. From there, metric tons of carbon can be arrived at by dividing gallons burned by 2204.6.

Now the hard one. Calculating the electricity consumed by a computer, and thus a traditional server, is best accomplished by attaching an electric meter to it. But we don't use traditional servers. We, like most providers, use virtual private servers (VPSes). A VPS is pure software that may share the resources provided by a physical machine with other VPSes. The electricity it requires is therefore some portion of the electricity required by its physical host. If the wattage used by the physical host and the number of VPSes residing on it is known and if each VPS is allocated similar resources, a rough approximation of the electricity required by each VPS can be calculated by dividing the host's wattage by the number of its VPS tenants. But such calculations are generally too simplistic to be accurate. VPSes can be moved to different hosts and tenants can be added, removed, or allocated different resources. The relationship of a host to any given VPS can therefore be fluid and tenuous. Unless the specifics of a VPS, its fellow tenants, and its host are known, and unless changes to any of these factors are followed, the electricity the VPS requires is difficult to calculate.

At iBec, however, the VPSes that we use are each single instances on their host machine, meaning that they do not share the machine with other tenants. This means that the electricity required by the VPS is roughly the same as the wattage required by its host. In our case, the host machine for each VPS draws roughly 600W of power. From there, it's not hard to get to the amount of carbon released: divide wattage by 1000 to get kilowatts, multiply by 24 and then 365 to get annual kilowatt hours, divide by 1000 to get annual megawatt hours, multiply by pounds of carbon released per megawatt hour according to https://www.eia.gov/electricity/state/index.php (it helps to know the state in which the server resides), and divide by 2204.6 to get the amount of carbon burned per year in metric tons (mT). This gives us a pretty good idea of the amount of carbon produced as a result of the electricity used by a single machine.

The calculation doesnâ€™t end there though. Each of these machines is part of a datacenter with its own electrical power demands, the most salient of which is cooling. To account for these additional demands, I used a PUE (power usage effectiveness) of two, meaning that the power required by a single machine is multiplied by two to account for the power overhead of the datacenter. With this final number in hand, along with the results of the gas usage and commute calculations, we can proceed to shop for an appropriate number of carbon offsets in order to balance our impact on the earthâ€™s atmosphere.