WirelessCar Achieves 10x Faster Startup Performance Using AWS Lambda SnapStart for Java

Founded in 1999, WirelessCar designs, develops, and operates connected car services for major car manufacturers. Its offerings include call center services, subscription management, and smart routing for electric vehicles. WirelessCar’s digital footprint is massive: the company’s services are used in more than 10 million cars in over 100 markets worldwide.

In an industry where instant connectivity and prompt responses are expectations rather than benefits, cold starts—the delay experienced when a serverless function handles its first request after a period of unuse—can have a major impact on the user experience. When Lambda SnapStart was announced, teams within WirelessCar realized its potential to improve startup performance.

“With Lambda SnapStart, you can take a snapshot from an already-initiated version of an AWS Lambda function to start it again,” says Fredrik Werna, developer for the Master Data team at WirelessCar. “You don’t have to spin everything up for the function to work, and therefore, you can save a lot of time.”

Within WirelessCar, the Master Data team was one of the first to implement Lambda SnapStart. This team manages a downstream backend service that ingests calls coming from its connected vehicle services and relies heavily on AWS Lambda functions. In March 2023, the Master Data team started a proof of concept (POC) to assess the performance of Lambda SnapStart within its environment.

The implementation process was simple; the service is fully managed and does not require significant setup or configuration time. The Master Data team needed to make only minor edits to templates in AWS CloudFormation, a service that lets developers model, provision, and manage AWS and third-party resources by treating infrastructure as code, to start running the POC. “We didn’t need any AWS support,” says Werna. “Because of the comprehensive guidelines, the implementation process for Lambda SnapStart was smooth and efficient.”

The POC’s results were promising. After evaluating a group of AWS Lambda functions, the Master Data team saw a 90 percent reduction in initialization duration. This amounts to a 10-times improvement in startup performance. “Before we implemented AWS Lambda SnapStart for Java, we had an initialization duration of approximately 4 seconds,” says Werna. “Now, it’s 0.4 seconds. The POC results made us very confident, so we implemented the feature in the rest of our production environment.”

The Master Data team now uses Lambda SnapStart for all 134 AWS Lambda functions in Europe and 121 functions in North America. Although the feature is not yet available in China, the team was able to implement conditions within its AWS CloudFormation templates to continue to deploy its AWS Lambda functions in that region. All in all, it deployed Lambda SnapStart with virtually no service interruptions or downtime.

After the Master Data team’s implementation, many other groups across WirelessCar adopted the feature to support their individual use cases. They enjoyed equally smooth implementation and performance benefits without incurring any extra costs; one team saw a 2-second reduction in cold-start times. “Implementing Lambda SnapStart was like flipping a switch,” says Anders Hultstein, enterprise architect at WirelessCar.