< Back to previous page


Load balancing and scheduling in large-scale computer systems.

Since the introduction of the very first communication networks, queueing models have played a key role in improving network performance. This has resulted in a large body of queueing theory literature that has found widespread use in many other areas of science and technology. As the area of computer systems and networks is ever evolving, so is the need for new, tailored queueing models. Large-scale systems (e.g., grid computing or cloud computing) have become quite prevalent today and are often composed of many heterogeneous resources. The analysis of such large-scale heterogeneous systems using traditional queueing theory is prohibitively expensive as the required time and memory complexity tends to scale poorly in the system size. The aim of this project is to introduce and analyze new queueing models that provide insight into the performance of existing and novel load balancing and scheduling algorithms for large-scale systems. The problems under consideration include affinity scheduling problems motivated by MapReduce clusters, load balancers that make use of redundancy to mitigate latency caused by server unpredictability, and stateful load balancers. The main envisioned methodology exists in developing fluid approximations that are validated using simulation experiments and that can be shown to become exact as the system size tends to infinity. The project combines techniques from stochastic modelling, probability, dynamical systems, numerical analysis and simulation.
Date:1 Jan 2019 →  Today
Disciplines:Systems theory, control