Workload-Aware Dynamic Load Balancer for Cold-Start Reduction in Serverless Function-as-a-Service Environments

Abstract

Function as-a-Service (FaaS) is serverless cloud computing model that automatically manage servers based on user requirements. However, due to this advancement there is problem called “Cold Start Latency” occurs. The cold-start latency bottleneck arises because FaaS send request to server without knowing server state i.e., warm or cold. Previous Load Balancer (LB) research has not mention functional instance life cycle and resulted in inappropriate routing of functional instances, increased latency, and increased costs. This work provides a detailed comparison of previous static LB and traditional dynamic LB models with a new state-aware instance allocation method, Workload-Aware Dynamic Load Balancer, (WADLB). Proposed algorithm uses computer simulation technique like pareto distribution to know how the system handles the request that comes in bursts not evenly, this helped to study how the system balances user delay (latency) and cost of operation under different traffic loads. Results shows that 65% decrease in request latency and reduce cold-start by keeping containers instances warm also increase in response time and reducing cost of resources in modern serverless environment. It enhances Quality of Service (QoS) by providing more consistence and predictable performance for end user.