Weighing clouds: how to assess computing infrastructure emissions

This is our third post in a series of four articles which dissect how to measure greenhouse gas emissions from operating data centers. Today’s post explores how to actually go about evaluating your organization’s emissions, including those from on-prem and hybrid infrastructure. 

Check out other posts in this series to learn about the primary sources of greenhouse gas emissions in data centers, the role of AWS, Azure, and GCP, and how you can lower your cloud computing footprint.

‍

To accurately assess greenhouse gas emissions for your organization, you have to look beyond your immediate environment. Every activity along your entire supply chain needs to be taken into account, and every business activity—including servers that may be hosting your applications and data, whether on-premises or in the cloud in an entirely different country—should be evaluated. This can be a daunting and complex exercise, but at Climatiq, we’re here to support. We’ve got a whole series of articles and resources dedicated to helping you understand how to go about estimating emissions from various business activities. But for now, let’s focus on assessing emissions from computing activities including cloud, on-premise, and hybrid setups. 

‍

How do I assess emissions from my cloud computing activities?

If you’re hosting your computing activities in the cloud, there’s a high likelihood you’re using one of ‘the big 3’ cloud providers—namely AWS, Azure, or GCP. Climatiq provides emission factors for each of them, but let’s sum up the basics here before you get into the nitty-gritty. 

To measure your cloud computing footprint you first need to: 

• Select a cloud provider
• Select the specific data center you’re using

The steps that follow are slightly trickier, depending on what you’re looking to measure. 

How to measure emissions from virtual machines

To measure VM (virtual machine) usage, you’ll need to understand CPU count, and the amount of vCPUs the machine has. This information is best pulled from the APIs of your cloud provider. If you’re not sure what the data center utilization rate is, 50% is a good fallback. You’ll also have to request info about allocations of disk and memory, and use the CPU, memory, and storage endpoints in tandem to get a full picture. Alternatively, you can define the model of virtual machine in use and then consult a table to find out how many vCPUs, memory, etc. this particular machine has. If the vendor doesn’t provide this data, this is a good reference guide. 

‍

How to measure emissions from cloud storage services

If you’re measuring emissions from cloud storage services, you’ll need to take a slightly different approach. There are 3 key things you need to know: 

1. Disk type: Is your data saved on an SSD disk or a HDD disk? You might need to guess if you’re using a managed storage service as the cloud provider won’t always declare this information.
2. Disk allocation: How much storage is allocated to these disks? If your cloud provider has a mounted disk, they should state how much storage you have available. If you’re using managed services where you most likely share hard drives, you should calculate for the amount of storage you’re actually using. 
3. Replication factor: If you’re using a managed service, what’s the replication factor? For example, AWS S3 replicates the storage in 3 locations so you’re actually using 3 times the storage you think you are because it’s shared across 3 different places. Be aware that replication factors are hard to find as cloud providers don’t always explicitly declare them. 

‍

How to calculate cloud computing CO2e beyond the big three

As we mentioned, Climatiq provides emission factors for AWS, Azure, and GCP, but does not currently provide data on other cloud providers. To estimate general cloud carbon emissions from other providers, Cloud Carbon Footprint has a good methodology you can use. For operational emissions, you’ll need to calculate the following: 

‍

‍

Your operational emissions will then need to be added to your embodied emissions. Embodied emissions are the estimated metric tons of CO2e generated by the manufacturing of data center servers for computing usage. These calculations are not included in Climatiq data right now but here is a rundown of how to go about calculating each factor: 

• Cloud provider service usage: Calculate the average vCPU utilisation, whether variable or constant. This can be pulled from your cloud provider’s API or, if they don’t offer that data, you can use 50% as a fallback.  
• Cloud energy conversion factors: The following is derived from an Etsy model:

​​— 2.10 Wh per vCPUh [Server]

— 0.89 Wh/TBh for HDD storage [Storage]

— 1.52 Wh/TBh for SSD storage [Storage]

• vCPU hours (variable): You can pull this data from your cloud’s usage APIs or billing data.
• Power usage effectiveness: The PUE factor for your local data center should be reported by your cloud provider.

• Grid emission factors: These can be sourced from Climatiq. 

You can find a table of embodied emissions for popular server models here, and also here, with a full list available here on Github. This should allow you to model the carbon intensity of a given cloud data center, although you may be restricted by how transparent—or opaque!—your provider is. 

‍

How to assess greenhouse gas emissions for on-prem or hybrid computing infrastructure

For on-prem or hybrid set-ups, the general methodology is similar, and CCF has, again, provided a detailed methodology here. The key to assessing your emissions is in understanding:

• Power consumption per server: Ideally, to get an accurate calculation you’ll want to measure power consumption per server. Some solutions like Dynatrace or Sunbird provide this data. Alternatively, you can use a wattmeter or sample some processes. The SPECpower database also has some median wattage values, but they are only reliable when you run a full server.  
• CPU utilization rates: These can be obtained from your server monitoring software. 
• PUE: The average PUE defaults to 1.58, based on this Uptime Institute Report. This can vary, of course, as older data centers tend to have a higher PUE. 
• Data center location and local grid mix: Data centers located in fossil-fuel regions will emit more carbon than those powered by renewables. Grid emission factors provided by Climatiq can be used for this. 

It’s important to note that these calculations exclude carbon emissions caused by water usage, especially wastewater treatment, which will differ depending on the geographic location. More comprehensive research and data are needed to properly factor this in when estimating the full carbon intensity of computing activities.

We hope this gives you a solid overview of how to accurately calculate emissions from your computing activities, whether that’s fully in the cloud or using an on-prem or hybrid set-up.