Open-source Serverless Architectures: an Evaluation of Apache OpenWhisk

Abstract

The serverless computing paradigm ushers in new concepts for running applications and services in the cloud. Currently, commercial solutions dominate the market, though open-source solutions do exist. As a consequence of this, there is little research detailing how well the different opensource solutions perform. In this paper, one such open-source solution, Apache OpenWhisk, is investigated to shed light on the capabilities and limitations inherent of such serverless computing architecture, and principally to provide further research on this particular solution's performance. This is accomplished through an extensive evaluation of OpenWhisk, involving a variety of experiments and benchmarks.

Performance Analysis of Apache OpenWhisk Across the Edge-Cloud Continuum

“…Maissen et al [15] studied the influence of request rate, cloud location, memory size, and programming language on latency across different serverless providers. Djemame et al [16] assessed OpenWhisk's effectiveness and efficiency on the cloud, comparing it to Docker and native function execution solutions. Back et al [17] and Lee et al [18] performed comparative analyses of OpenWhisk, AWS Lambda, Google Cloud, and Microsoft Azure, considering aspects like execution time, cost, and resource utilization.…”

Let's Trace It: Fine-Grained Serverless Benchmarking using Synchronous and Asynchronous Orchestrated Applications

“…Over 100 studies from academia and industry have already appeared [28,59,63,88]. Commonly investigated topics include scalability [37,49], cold starts [48,81], performance variability [15,38], instance recycling times [43,81], and the impact of parameters such as memory size [24,89], or programming language [20,31]. These studies tend to rely on single-purpose micro-benchmarks and rarely utilize tracing data.…”

On the Power Consumption of Serverless Functions: An Evaluation of OpenFaaS

“…In a computer system, the CPU is considered to be the first consumer of power and is followed by the memory [15]. The CPU and the memory are therefore the most important resources to be evaluated in terms of performance in the context of applications execution [16]. Consequently, they are also important resources to consider when measuring power consumption.…”

“…Multiplication: This benchmarking application consists of the multiplication of two square matrices with a size of 5000. The application is used to stress the memory more than any other hardware component [16]. There are two versions of this benchmark that were written to run as an OpenFaaS function and a Docker container, the source code of which can be found in [23] under the names "Matrices-Function-Benchmark.zip" and "Matrices-Docker-Benchmark.py".…”