Low Latency Warmup Time For Real Time Applications In Serverless Computing

Abstract: The Serverless computing is new paradigm in cloud-native applications. The cloud-native applications are Resilient, Highly Available and Scalable, and achieved by CPU and/or Memory usage metrics. Scale-to-zero is to bring down application which is idle i.e., when it is not processing any request or where there is no computing. The standard benchmark CPU/Memory metrics would never be zero and hence not helpful in determining whether application is idle or not. Hence custom metrics would be defined and assessed to bring down the application by determining its idle state. On new request, application new instance boots up and process the request. This booting time can be seriously affecting the latency and throughput for requests which are real time in nature. To improve the responsiveness for real time applications, scale-to-near-zero can be incorporated. Scale-to-near-zero is where, small footprint application instance exists always, which is capable enough to process real-time request. This improves the responsiveness of real time applications, eliminates the warmup time, and maximizes chances of being deterministic. Full capacity application instance can be instantiated based on the current load of small footprint application. The discussed approach targeted for FaaS (Functions-As-A-Service) of serverless computing.

Patent Information

Applicants

Inventors

Specification

Documents

Application Documents