Low emissions Search.

“Be the change you want to see in the world.” — Author disputed

Introduction

The majority of climate scientists around the world believe that greenhouse gas emissions are directly related to climate change. The IPCC report calls for immediate action to curb greenhouse gas emissions to mitigate the effects of climate change. Software organizations who run large scale cloud infrastructure can do their part too in addressing climate change. In this quick read, we examine how my current team reduced its operational carbon footprint by approximately 60% using AWS Graviton-2 hosts that deliver the same performance at 1/3 the power usage.

Search and Low Emissions

Data centers are one of the highest consumers of power per square foot across any commercial establishment. It is estimated that they consume up to 2% of the energy produced in the United States [1]. In figures, it’s estimated[2] that this is north of 205 Terawatt hour. On average, a kWh of energy produced from coal or petroleum, causes 2 pounds of CO2 Emissions. When you convert that to 205 TWh, we are looking at a 2 X 2.05e+11 pounds of CO2 emissions.

Search (Information Retrieval) by nature is computationally intensive — search servers typically need CPUs with plenty of cores and fast network and I/O to provide a fast search experience. As a result of this search clusters are power hungry and are linked to higher emissions. My team at work was aware of this and was investigating ways to reduce the emissions. The initial approach was to perform on-demand scaling of infrastructure. For e.g. when the team expected a higher usage like a tax filing deadline we would prescale the infrastructure and scale it down post spike. This worked, but the team knew that the real solution will be the availability and adoption of more power efficient computing.

In May 2021, AWS announced that it was offering its Graviton-2 processors for AWS managed OpenSearch. Graviton-2 processors are the most power-efficient processors that AWS has deployed to date. These processors provide 2 to 3 times the computing power per watt consumed[3][4] compared to a comparable M* instance.

This opened up the ability to leverage Graviton hosts to run our clusters both for computational efficiency and cost savings. Partnering with AWS, my team started to migrate the OpenSearch clusters from non-graviton instances to Graviton.

As the first step, we migrated a single non-production instance to Graviton and let it go through the standard performance and endurance tests. After approximately a week of testing, the team compared the performance and endurance benchmarks with the existing infrastructure. We found that the benchmarks were meeting or exceeding our expectations.

The second stage was to migrate the internal customer test environments to Graviton. This took around 1.5 week of time.

Finally, the team migrated the production clusters to Graviton-based hosts one by one starting with the smallest cluster working up to the largest and mission critical ones. The production migration effort took around 1 month to complete. By the end of approximately 2 months all IPS Search cluster hosts were running Graviton-2.

Today all of the clusters in operation run on Graviton-2 and is the default host configuration that we use to provision search infrastructure. This transition has helped reduce our carbon footprint by ~60% and reduced our cloud spending by 20%. We are also in the process of migrating the ingest and read infrastructure to Graviton based hosts. The team has never encountered any performance or stability issues related to the migration over the past one year since we did the migration.

AWS has provided a seamless way to migrate your existing non Graviton clusters to graviton either from the console or using its supported Infrastructure As Code tools.

To summarize, we are proud of the fact that we did our (albeit small) part to reduce our operational carbon footprint and hope that this will inspire other software teams and engineers who come across this journey to make the right choices for a sustainable future.

Note of thanks to Agil John from AWS who provided me all the reference materials around AWS Graviton.