Improving Processing Speeds up to 50% Using AWS Graviton Instances with HERE Technologies

With over 35 years of experience, HERE offers location technology to customers in more than 200 countries around the world. Its Tour Planning API is a versatile route planning and optimization tool that creates dynamic shift planning for last-mile fleets and minimizes cost or distance traveled based on available vehicles and drivers. It also maximizes throughput and deliveries at the lowest cost and is simple to use with existing in-house solutions and data. Customers can optimize routes based on parameters such as shortest time, shortest distance, or lowest cost. Tour Planning also uses a wide variety of additional route-optimization constraints, such as route restrictions based on near-real-time traffic data, to identify options for further cost and efficiency savings.

The final leg of a delivery, or “the last mile,” from a local hub to the customer’s home is considered the most expensive. Optimizing that route can greatly affect cost, customer experience, and environmental impact. To help customers with this challenge, HERE launched Tour Planning in 2020 using compute-optimized Amazon EC2 instances that run in Kubernetes clusters on the HERE Delivery Platform, a shared infrastructure with 18,000 deployments. Throughout the project, HERE received support from AWS representatives, meeting weekly to troubleshoot and strategize. With a focus on cost optimization, HERE participated in a deep-dive session in 2022 on AWS Batch, a fully managed batch computing service for running batch processing, machine learning model training, and analysis at nearly any scale. As part of the session, AWS introduced HERE to AWS Graviton processor, which delivers optimal price performance for cloud workloads running in Amazon EC2.

Figure 1. HERE Tour Planning architecture diagram

In June 2022, the Tour Planning team began migrating its asynchronous infrastructure, which runs complex problems that take longer to yield results, to AWS Graviton–based instances. For 6 months, the team tested performance, availability, and cost impact in nonproduction workloads. Satisfied with the results, the team migrated its asynchronous production workloads to AWS Graviton–based instances in January 2023. Using AWS Batch and AWS Graviton–based instances for asynchronous workloads, HERE reduced costs by up to 30 percent per request while also increasing performance.

In January 2023, HERE’s asynchronous workloads started using Amazon EC2 Spot Instances, which let users take advantage of unused Amazon EC2 capacity on AWS. HERE increased its use of Amazon EC2 Spot Instances on research and development deployments from 45 percent in 2022 to 88 percent by August 2023. This saved HERE up to 55 percent for its most common usage profiles. With these cost savings, HERE can offer the same service and performance to customers at a lower cost or use the savings to offer more features, capabilities, and performance to customers at the same price as before. “When we save on costs, we can be more competitive in terms of pricing for our customers,” says Suresh Gupta, engineering manager for Tour Planning at HERE. “Or we can help them achieve even better results within the same runtime by factoring in more route-optimization variables.”

After a lunch-and-learn session with a specialist for AWS Graviton–based instances in April 2023, the Tour Planning team began migrating its synchronous workloads—smaller, simpler problems that run in seconds and account for 90 percent of Tour Planning’s API calls. Due to the robust testing HERE had done, the actual migration to AWS Graviton–based instances for the synchronous workloads took 1 week with zero downtime. “Usually, when we change part of our infrastructure, issues pop up, and we have to make fixes or adjustments later,” says Tatu Niemi, lead DevOps engineer for the Tour Planning team at HERE. “Using AWS Graviton–based instances, we had no issues. It worked from the start.”

HERE also improved processing speeds for synchronous workloads using AWS Graviton–based instances. “Fast response time matters more for synchronous workloads, and we saw up to 50 percent faster responses and sometimes even more,” says Niemi. “For smaller optimization problems specifically, we see notably fast results.” HERE also doesn’t need to scale up as frequently or as high as before; it spins up larger-than-usual numbers of instances during peak hours 60 percent less often. HERE estimates that they are saving 10–15 percent for synchronous workloads because instances are less costly and the company needs to use them less frequently.

HERE aims to have net-zero carbon emissions by 2035. AWS Graviton–based instances use up to 60 percent less energy, which supports HERE’s sustainability goals. “Using AWS services, HERE scales up elastic compute activity less often, using less energy and making our environmental footprint smaller,” says Niemi. The company also increases sustainability by using fewer instances for Tour Planning. In addition, HERE helps its customers meet sustainability goals by minimizing distances traveled and avoiding traffic congestion, which helps to reduce carbon emissions.