Skip to main content

Master Thesis Projects

Meroton offers opportunities for master theses work. As a master thesis student, you will sit together with our team in our Linköping office and get daily direct support from a mentor on your task.

This page lists thesis subjects that we have identified as interesting and suitable. Alternatively, you are welcome to propose your own subject.

Contact Person​

Fredrik Medley <fredrik@meroton.com>
Telephone: +46730739188.

Predictive Scaling of Dynamic Cloud Workloads​

Background​

The IT industry typically scales workloads using discrete, ad-hoc logic based on resource utilization. While effective for simple cases, these methods are often suboptimal for highly dynamic workloads, like those in large build clusters.

Highly dynamic workloads often exhibit a predictable underlying signal that simple heuristic methods fail to exploit. This thesis project aims to apply modern control theory to leverage these signals for efficiency gains.

Task​

The thesis will investigate and quantify the potential savings from an optimized scaling algorithm. The core task is to develop and evaluate several general models such as: PID controllers, model predictive control (MPC), and reinforcement learning (RL) approaches.

The models will be evaluated with real world data. A key part of the analysis will be comparing the trade-offs between critical performance indicators like queue sizes, resource overallocation, as well as the operational effort required from cluster maintainers to keep the system tuned.

Prerequisites​

  • Student in a Master's program for Computer Science, Control Engineering, Electrical Engineering, or a related discipline.
  • Strong foundation in control theory (PID, MPC) and/or machine learning.
  • An analytical mindset with the ability to apply theoretical concepts to real-world problems.
  • Preferred: Golang and Kubernetes knowledge

Reduce Build Times by Trimming Input File Trees​

Background​

Large scale development projects often suffer slow continuous testing experience. A build tool like Make runs lots of actions, e.g. compiling a file, linking a binary or executing a test. Each action can be described by its input file tree and command line. If either the file tree or the command line changes, the action has to be rerun.

The input file tree has a tendency to be overspecified, for example using a whole Python package when only a few files are imported. This leads to rerunning actions when the modified files are not used.

Task​

This thesis project will investigate the potential savings from optimising the input file trees to the build steps, based on the used files, reported by a virtual file system (FUSE), instead of the usually overspecified input file lists. For evaluation, a couple of open source projects will be selected and evaluated. The task will include Go programming for extracting data and then analysis using languages of your choice.

Prerequisites​

  • Required: Linux programming skills.
  • Preferred: Golang knowledge.