CalcsService | Calcs service for Severless Application Analytics Framework

CalcsService is a Shell library. CalcsService has no bugs, it has no vulnerabilities and it has low support. You can download it from GitHub.

The CalcsService is a compute-bound microservice implemented for testing Function-as-a-Service (FaaS) platforms. Implementations are available in Java, Node.JS, and Python. The CalcsService integrates the Serverless Application Analytics Framework (SAAF) to obtain CPU and performance metrics. The CalcsService implements random multiplication and division where the operands are stored in arrays for math operations as opposed to storing operands using primitive variables. Array indexes are selected randomly forcing the application to jump large distances across the arrays to read the operands increasing accesses to different memory pages to effectively negate system page caches. Memory stress can then be incremently introduced by using increasingly larger arrays. Using this technique we demonstrate an increase in the number of page faults and a resulting increase in runtime for the CalcsService. Using an array size of 1,000,000 vs. 1 increased the number of page faults 15.8x measured using an AWS Lambda function deployed with 256MB. With a 2048 MB function function deployment, the number of page faults increased 13.2x. Page faults are shown to produce increased function runtimes on AWS Lambda of 3.9x (256MB), 3.88x (512MB), 3.55x (1024MB), and 2.24x (2048MB) respectively. Increasing the array size also increases performance variance of the function presented here as the coefficient of variation (CV) from 11.9% -> 29.7% (256MB), 9.6% -> 25.8% (512MB), 9.2%->21.7% (1024MB), and 5.5%->24.8% (2048MB). The measurements were derived using the Java version of the Calcs Service. Our benchmarks with SAAF only consider the actual runtime of the function code. We have removed network latency and infrastructure cold start latency time and report the impact on the actual calculation runtime removing FaaS platform overhead.