Zwift Improves Single Sign-On DataGrid Reducing Latency by 60%

Download Case Study

60%

Reduced Latency (P95)

14%

Reduced Latency (P75)

25%

Reduced CPU Utilization

Zwift is determined to create an optimal experience for players using its live fitness app, and as part of its efforts the engineering team has largely focused on creating and maintaining an infrastructure that supports high-level performance. After a short PoC with Granulate, Zwift was able to achieve a significant reduction in both latency and CPU utilization on its Single Sign-On DataGrid service, with no engineering resources.

About

Zwift is a one-of-a-kind fitness company born from gaming, with a unique app that makes indoor training fun by connecting cyclists and runners around the world. The Zwift app mixes the intensity of training with the immersiveness and engagement of gaming, allowing its users to ride or run in virtual worlds with a community that motivates them to keep going. A key component of the experience is the massively multiplayer online gaming model, plus the technology to create rich, 3D worlds ripe for exploration.

To ensure the game world runs smoothly for all active users, no matter how many are currently connected, Zwift hosts its app in AWS EKS environments.

Since 2015, Zwift is a paid app with a growing user base. Players may cycle freely around the game world and join organized group rides, races, or workouts with other users. The app connects wirelessly to exercise equipment: bike trainers, treadmills, and more, so real world effort powers the player's avatar in the game, by converting it to speed and power.

Granulate was not only easy to deploy, but we were also excited about the results with 60% reduction in latency.

Roberto Duarte, VP Engineering

The Challenge

As a gaming and training app, it is central to Zwift’s success that the company provides an optimal user experience by maintaining strict performance requirements. Being designed around community and real-time engagement, the service must run smoothly no matter how much player traffic or server load there is at any moment.

There are also multiple data inputs and outputs within the Zwift ecosystem – to and from the exercise devices, live rendering of the game world, real-time progress tracking and more – making low latency a primary goal.

Zwift’s team was seeking a complementary solution to previous optimization efforts, where a substantial improvement could be achieved without requiring code changes or R&D overhead.

The Results

Following the activation of Granulate’s agent, Zwift saw an improvement of 60% in P95 latency and a 14% reduction for its P75 in some of the calls involved in the customer authentication into the single-sign-on service. They also achieved 20% reduction in CPU utilization, providing them with headroom to support future growth.