The fundamental goal of my research is to gain an understanding of existing strategies for scheduling large computational tasks in modern high-performance computing (HPC) environments and then build on this to eventually design, implement and evaluate novel ones.
Modern HPC architectures are increasingly heterogeneous (for example, with multiple CPUs and GPUs), so one of the major current challenges is to ensure that tasks are assigned to the resources best suited to handle them. Effective scheduling strategies for heterogeneous architectures are therefore vital.
One of the aims of my work is to attempt to estimate an optimal scheduling strategy on a heterogeneous system by applying techniques from reinforcement learning that have recently proven successful for similar problems in other contexts.
- Numerical analysis.
- High performance computing.
- Numerical algorithms and software.
- My LinkedIn profile.
- I am currently the Treasurer of the Manchester SIAM-IMA Student Chapter. Sign up to join us!
- I am a member of the Numerical Linear Algebra Group at Manchester.
- I completed an Applied Mathematics MSc at Manchester in September, 2017. My thesis was sponsored by the Numerical Algorithms Group and concerned the modified Cholesky decomposition for symmetric indefinite matrices and its applications, which include computing a bound on the distance to the nearest correlation matrix. This can be found here.
Email: thomas.mcsweeney [at] postgrad.manchester.ac.uk