Estimating Multiclass Service Demand Distributions Using Markovian Arrival Processes

Building performance models for software services in DevOps is costly and error-prone. Accurate service demand distribution estimation critical to precisely modeling queueing behaviors prediction. However, current methods focus on capturing the mean demand, disregarding higher-order moments of that still can largely affect prediction accuracy. To address this limitation, we propose estimate higher a microservice from monitoring traces. We first generate closed model abstract use it departure process requests completed by as Markovian arrival (MAP). This allows formulating into an optimization problem, which aims find multiple maximize likelihood MAP using generated measured inter-departure times. then different classes with maximum algorithm novel heuristics mitigate computational cost scalability. apply our method real traces microservice-based application demonstrate its estimations lead greater accuracy than exponential distributions assumed traditional approaches services.