A Novel Function Fusion Approach for Serverless Cold Start

Serverless computing is a new cloud technology that is popular due to its fast reaction time and scale-to-zero features, where users are only charged for the actual quantity of resource utilization and the code can work closer to the users to minimize the delay. The code that can run independently on a different container serves as the processing element in serverless computing. A container needs to be set up and prepared each time a request for function implementation is made. This is referred to as a "cold start," and it takes some time before the actual execution starts.The cold start issue substantially impacts the overall response time of a workflow that consists of functions because it can happen in any of the functions. One method for reducing a workflow's cold start delay is function fusion. When two functions are combined to form one function, the cold start of the second function is removed. However, if parallel functions are fused, the workflow reaction time can be increased because the parallel functions run sequentially even though the second function's cold start is eliminated. This study provides a method for using function fusion and taking into account a parallel run to reduce the cold start latency of a workflow. We first identify the three latencies that have an impact on response time, then we show a workflow response time model that takes the latency into account. Finally, we effectively locate a fusion solution that can improve response time on a cold start.