Learn how Xerris's team of AWS experts and Kubernetes Certified Administrators created a cost-effective and cloud-agnostic real-time chat infrastructure that could scale to 47,000 transactions per second.
Two Hat was looking for an AWS partner with Kubernetes certified administrators on the team to complete a detailed design for a cloud-agnostic Kubernetes implementation, running in AWS. The solution Two Hat was looking for needed to be able to support 47,000+ transactions per second and have a response time of fewer than 30 milliseconds to satisfy the real-time chat needs of Two Hat.
From the get-go, Xerris suggested a quick assessment that included several days of workshops with Two Hat to understand their principal architect’s long-term vision for the Kubernetes implementation. Xerris’s team then completed an implementation roadmap that addressed network level, service mesh questions, and traffic shaping. With day 1 and beyond operations in mind, Xerris also made a recommendation to implement AWS Game Day scenario training into the weekly sprint schedule.
A solution Xerris proposed would work with AWS EKS and Terraform for the infrastructure code solution to be implemented. Focusing on Terraform leveraged Hashicorp’s templates of AWS providers to ensure scalability but also configurations that made sense for Two Hat at the time.
Using AWS EKS made it easier for the client to maintain a fleet of spot and reserved node groups. This allowed the Two Hat team of developers to switch focus to learning how to effectively monitor the solution using Prometheus, and Grafana. Focusing on using GitOps by implementing a tool like Flux allowed the team to help increase the ease of deploying to multiple clusters as well as giving auditability to the whole process.
The result - using Kubernetes reduced Two Hat’s EC2 costs significantly and their deployment to market times have been reduced from several hours to minutes. The scalability of Kubernetes using horizontal pod auto scalers allows the application to scale up when demand is high and scale down when fewer clients using their service.
Premeditated Game Day training and monitoring tools like Prometheus reduced stress on the Two Hat Operations team and allowed them to manage alerts and respond to production issues quickly as a team.
transactions per second
milliseconds response time
Release cycle deployments in minutes, not hours