Time-to-market and engineering efficiency are the most critical and expensive metrics for a chip design company. With this in mind, the team at Annapurna Labs selected Altair Accelerator™ job scheduler for their front-end and back-end workflows. The team was managing workloads on a number of dedicated Amazon Elastic Compute Cloud (EC2) instances and they could scale up by manually adding new On-Demand instances. However, the process was not fully automated and Annapurna Labs used Rapid Scaling to add structure and efficiency to scaling AWS compute resources, shorten time to results, and change the development model to Continuous Integration.
In addition to automatically starting new instances when there is demand, Rapid Scaling looks at the speed at which the demand is being met and stops scaling up if the speed is good enough. This means demand can be met in 10 minutes. The license-first approach to scheduling allows Accelerator to efficiently differentiate workloads waiting for licenses versus workloads waiting for hardware. All resources are freed after they have been idle for one minute.
Altair worked with Annapurna Labs to add more features including configurable selection of instance types, Spot Instance support, protection against various errors like saturation of instance types, size of /tmp, fine control of the number of jobs that can be executed on each new instance, and many others.
Electronic design automation (EDA) jobs can be short and the spin-up time for an instance is comparable to those jobs’ runtime. The ability to understand workload speed and spin-up costs enables Rapid Scaling to avoid overshoot. Amazon EC2 offers the broadest and deepest choice of instances, built on the latest compute, storage, and networking technologies and engineered for high performance and security. Rapid Scaling allows job resource requests to map to the most appropriate instances. Rapid Scaling understands how to select a backup instance type if the first choice is not available. After the workload surge has passed, idle instances terminate. This flexibility maps nicely into AWS notions of Reserved, On-demand, and Spot instances.
Rapid Scaling allowed Annapurna Labs to reduce costs by at least 50%. Additionally, with Rapid Scaling now part of Annapurna Labs’s chip development Continuous Integration flow, they are seeing faster incremental development and continuous regression. Annapurna Labs keeps tighter control on costs and benefits from a detailed view into resource usage by projects and users.
Read full article here