A framework to compare how serverless systems scale resources

Researchers developed a unified modeling framework to evaluate different autoscaling strategies under bursty workloads.

A unified queueing-based framework was developed to compare autoscaling strategies. This framework evaluates how serverless platforms handle bursty and unpredictable workloads.

The work models three representative autoscaling strategies. These are scale-per-request (SPR), concurrency-level scaling (CLS), and metric-based scaling (MBS).

The framework combines analytical queueing network (QN) modeling with queueing Petri nets (QPN). It uses Java Modeling Tools (JMT) to evaluate behavior under two-phase hyper-exponential (H2) arrivals.

Evaluation uses four system-level metrics. These are throughput, mean response time, system population, and system power.

SPR provides the lowest delay under moderate load but becomes fragile near capacity limits. CLS is governed by concurrency tokens and is sensitive to configuration under bursty conditions.

MBS provides the most stable quality of service (QoS) across the workload range. It keeps utilization below a critical region to prevent explosive queue growth.

Key facts

  • The framework evaluates SPR, CLS, and MBS strategies.
  • Evaluation uses throughput, mean response time, system population, and system power.
  • MBS provides the most stable quality of service across the workload range.

All sections

Edition