Spotify’s infrastructure hits all the right performance notes with the help of Axion processors

*Featured in this video:* Dave Zolotusky, Principle Engineer at Spotify

*Executive summary:* Spotify, the world’s most popular audio streaming subscription service, counts on Google Cloud to help 788 million users, including more than 300 million subscribers, find and enjoy the right music, podcast, or audiobook for every moment. To maintain high delivery standards, Spotify migrated much of its backend services to Google Axion Processors, which are custom-built for Google Cloud to deliver unrivaled performance and energy efficiency. By adopting an Arm-based architecture, Spotify has significantly reduced the operational overhead of running the platform’s processing power — between 1.5 and 2 million compute cores — while delivering backend performance up to 2.5x as fast and as much as 40% reduction in compute costs.

*Challenge:* With millions of tracks and episodes, Spotify needs consistent performance, power, and scale to connect people across the globe with their favorite listening experiences on demand. The main challenge is ensuring its robust infrastructure can handle varying traffic peaks across different regions cost-effectively while advancing its commitment to sustainable, energy-efficient operations.

*Solution:* Building on its long-standing partnership with Google Cloud, Spotify decided to migrate its core backend services to Google Axion Processors — Google’s first custom Arm-based CPUs. Designed to deliver industry-leading price-performance for diverse workloads, Axion is helping to improve the efficiency of Spotify’s underlying architecture while providing the high availability and scalability needed to deliver the best listening experiences while also minimizing the time engineers have to spend dealing with upgrades, failures, and overall maintenance complexity.

*Results:* With Google Axion as its compute platform, Spotify is capturing more compute value for every dollar — achieving up to 2.5 times faster backend performance and as much as 40% reduction in compute costs. Engineering teams also benefit from effectively zero-cost latency gains, ensuring more reliable and responsive listening experiences without the operational overhead.

*Key takeaways and highlights from our interview with Dave Zolotusky, Principal Engineer at Spotify:*

→ “We were running somewhere between one and a half and two million compute cores. Every cluster upgrade, every node failure at that scale. It’s just a lot of work. It’s significantly less operational overhead if we move to a modern architecture like Axion.”

→ “We have significantly faster performance — two to two and half times faster than what we had previously and 40 plus percent price-performance savings. In the cloud, being faster helps our engineers. That saves you a bunch of time that lets you do more things.”

→ “There's a very clear correlation with end user happiness and the performance of Spotify. A bigger or faster VM or faster spin up time is great because if suddenly we have a moment where a huge part of the world opens up Spotify, then we need to very quickly respond. Axion certainly helps with that.”

→ “The migration to ARM is really just a very simple code change. Super smooth and easy, which is kind of a surprising thing when you're talking about migrating both huge numbers of compute cores, but also changing architectures.”

*Google Cloud products used:* Google Axion

*Learn more:*
→ Spotify and Google Cloud https://cloud.google.com/customers/spotify

→ Google Axion processors boost AlloyDB, Cloud SQL, major customers, and ISVs https://cloud.google.com/blog/products/compute/google-axion-powers-cloud-sql-and-alloydb

→ Google Cloud and Spotify expand partnership to help unlock creator potential and reach fans https://www.googlecloudpresscorner.com/2023-11-16-Google-Cloud-and-Spotify-Expand-Partnership-to-Help-Unlock-Creator-Potential-and-Reach-Fans