PhD Defence | Commodity Compute- and Data-transport System Design in Modern Large-scale Distributed Radio Telescopes
The study of the radio universe is very a very data- and compute-intensive science that relies critically on large compute and data-transport systems. An important design goal of such systems is to maximise the amount of scientific discovery for the investment made.
In this thesis, Chris Broekema introduces a model to express the cost, but more importantly also the value of such systems allowing more granular evaluation of various systems, based not just on their cost, but on their value potential as well. Broekema completed his research under the supervision of Henri Bal (VU) and Rob van Nieuwpoort (UvA).
Broekema argues that, since modern radio telescopes are generally capable of producing overwhelming volumes of data, the data-transport system in such an instrument should be architected and designed together with the compute infrastructure. Examples both in the current LOFAR telescope, as well as in the Square Kilometre Array still under development, show that this co-design of data-transport and compute systems has significant value benefits. When these systems are considered together, interesting optimisations on the boundary of the two may be considered. Broekema shows two such optimisations that focus on controlling data-flows and reducing energy consumption of such data-flows respectively.
Finally Broekema considers the future. Over the last couple of decades compute capacity has continued to increase both predictably and dramatically. Current manufacturing technologies are approaching physical limits, which make this trend unsustainable, at least for conventional technologies. He inventories possible alternative technologies and how these may be applied in support of a radio telescope.