Firmware Management for a Heterogeneous Platform

Abstract

Heterogeneous hardware, much like its conventional counter parts runs firmware and other low-level software like bitstreams in the case of FPGAs. Current firmware management solutions only focus on very specific hardware or do not focus on enough variety regarding their hardware support like FPGA programming. A lack of firmware management can lead to longer downtime, which in turn leads to significant profit losses in a commercial setting, something that should ideally be minimized. As a result, in this thesis, we propose a starting point for managing firmware for all kinds of different hardware by walking through the design process and presenting an implementation specific to Enzian machines. We then evaluate our approach by first qualitatively reviewing the subsequent implementation with the help of usability testing. The subsequent qualitative performance evaluation using the proposed firmware manager shows that by leveraging the distributed property of server clusters, we can lessen downtime by an upper bound of the amount of machines present in the network, which can save costs by an order of magnitude.