On the simulation of advanced computer networks

The High Performance Computing (HPC) services of today provide the computational density to perform groundbreaking science. These services facilitate the distribution of a problem space across numerous compute units, yielding performance improvements through parallelising the execution of a workload. Although the design of any one of the computational units in use can significantly impact the performance of a scientific workload, the network that provides the communication between them can serve as an equally disruptive bottleneck for a workload’s performance.

From the perspective of modern HPC networks, various topologies, routing logics, and advanced features, such as in-network computation, combine to form a multitude of options an architect or researcher must consider. Of the tooling used by architects and researchers to navigate this design space, simulators are commonplace and offer a powerful methodology for investigating the efficiencies and pitfalls of potential solutions.

During the PhD programme, a student would be expected to explore the design of advanced computer networks, both through simulation and the direct execution of workloads on real-world implementations. As part of such exploration, the procurement or development of relevant workloads that are representative of real-world behaviours impacted by the design decisions of the networks under study should also be carried out. Additionally, all aspects of simulation and the underlying creation of network models represent a non-trivial exercise. The learnings of such explorations may be applied through the refinement of theoretical network models, representative of possible next-generation HPC networks.

Alongside the University of Bristol's High Performance Computing research group, the PhD programme will be associated with the Bristol Centre for Supercomputing (BriCS), home to Isambard-AI. BriCS is delivering over 300 million pounds of AI compute, and a successful applicant will be able to work closely with the BriCS team and their collaborators while delivering world-class supercomputer services.