Performance Modelling and Optimization of Serverless Computing Platforms

• Author / Creator

  Mahmoudi, Nima

• During the past 10 years, we have witnessed the proliferation of cloud computing services and their adoption in the industry. This rapid growth has been mainly due to economies of scale, improving resource utilization, infinite computing resources on demand, and pay per use cost model. However, the abundance of cloud computing services and infrastructure has left us with lots of resources to manage, which could surpass the development effort for a product. Several new paradigms have emerged since the introduction of cloud computing services, which have tried to improve the services and address these shortcomings. The most promising paradigm shift for the newly emerging cloud services is serverless computing. As a result, we are currently amid an evolutionary paradigm shift in cloud computing towards serverless platforms. This change is due to several improvements over traditional cloud computing, like handling virtually all of the administrative tasks, improving resource utilization, potential operational cost savings, improved energy efficiency, and more straightforward application development. In addition, serverless computing can enable the rapid development of new cloud-based solutions in our current highly-dynamic environment, enabling new solutions for unforeseen circumstances, e.g., pandemic-era applications. However, the current implementations of serverless computing are far from perfect and have a long way to achieve the full potentials of this paradigm.

  This thesis focuses on modelling and improving different aspects of serverless computing platforms. As mentioned above, in serverless computing, a large portion of operational tasks are delegated to the cloud operator, which frees the user from these tasks but leaves the operator to create a management system able to handle almost any type of workload. The current generations of serverless computing have emerged to sub-optimal management solutions, leading to underutilized infrastructure, exaggerated costs, and unsatisfactory latencies violating most Quality of Service (QoS) guarantees. One of the major reasons behind these shortcomings is the fact that the current serverless computing management systems are workload-agnostic, i.e., they don't adapt to the type of workload being executed on them. Global effort and research is needed to develop management systems capable of running most types of workloads with near-optimal behaviour.

  This thesis strives to develop analytical and data-driven models and methods and leverage them to improve the status quo in current serverless computing platforms. We aim to develop several optimization modules using different techniques, complementing each other in different scenarios.

• Subjects / Keywords

  • Performance Modelling
  • Optimization
  • Cloud Computing
  • Serverless Computing
  • Machine Learning
  • Performance Simulation

• Graduation date

  Spring 2022

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-ffdb-h989

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.