Improving performance by 30% using AWS Graviton4 with Sprinklr

As an innovator, Sprinklr has always been eager to adopt new technologies. “Sprinklr’s philosophy is that we’ll always have an edge if we use the latest and greatest technology,” says Nitin Goyal, vice president of engineering at Sprinklr.

Sprinklr’s forward-thinking approach has paid off in the past: The company’s earlier migration to Graviton3-based instances resulted in significant gains in price performance. “When we migrated some instances from Graviton2 to Graviton3, we saw better latencies and throughput,” says Goyal. “This made us confident that migrating to Graviton4 would provide similar improvements.”

Sprinklr especially hoped to improve scale-up time for applications, including improvement of tail latencies—the delay experienced by those few requests with longer-than-average processing times. “Faster pod start-up time means that we’re more responsive, not only to the scale-ups but also to any unknown failures,” says Goyal. “This improves availability and resiliency.” Features that would especially benefit from improvements in response times include Sprinklr’s live chat and contact center capabilities, which require near real-time responsiveness during customer interactions.

Sprinklr has a container-based microservices architecture, with backend applications written in Java. The company runs these applications on Amazon Elastic Kubernetes Service (Amazon EKS) to streamline Kubernetes applications by automating cluster management, helping its teams to focus on innovation.

Because there were different types of workloads to migrate, Sprinklr organized its migration along three separate tracks. “For our Kubernetes or stateless workloads, we moved to Graviton4 as soon as it was launched,” says Goyal. “It was a no-brainer for us.” The migration of those workloads was completed within weeks. The second part of the migration involved Sprinklr’s x86-based workloads, which took approximately 3–6 months due to the timing of release cycles. “Because we have quarterly release cycles, after moving a workload to a different architecture, we make sure that they’re working fine by testing in our queue environments,” says Goyal.

The third part of the migration focused on Sprinklr’s databases, which required additional validation and compatibility checks. That migration is expected to be completed soon, as additional Graviton4 storage-optimized instances become generally available.

By migrating x86-, Graviton2-, and Graviton3-based instances to Graviton4-based Amazon EC2 R8g instances—which provide the best price performance for memory-intensive workloads in Amazon EC2—Sprinklr streamlined operations. “Moving more of our workloads to Graviton4-based instances helped us standardize everything, so it’s easier to manage and maintain our infrastructure,” says Goyal.

Sprinklr also achieved significant performance gains, which were especially noticeable in the live chat and contact center features, where faster scale-ups translate into quicker response times and better customer experiences. “With our adoption of Graviton4-based instances, we’ve seen that our existing services are able to scale up quickly, reducing the turnaround time for our customers’ end users,” says Goyal.

For Java-based microservices, Sprinklr experienced improvements of 15–25 percent depending on the workload. And for storage-intensive workloads, Sprinklr increased overall performance by 30 percent while reducing the number of cores by 20–30 percent. Tail latencies also improved significantly, leading to faster response times. “Getting these performance and throughput improvements as compared with our x86-based processors means that we can do more with less,” says Goyal.

Looking ahead, Sprinklr plans to further optimize its infrastructure by migrating additional workloads to Graviton4, including its AI workloads. The company remains committed to continuous improvement using AWS technologies to enhance performance, resiliency, and customer satisfaction. “We want to keep improving the availability of our platform, reducing the points of failure and scaling customer workloads,” says Goyal. “It’s a continuous loop of innovation.”