On the use of heterogenous computing in high-energy particle physics at the ATLAS detector

Abstract

The ATLAS detector at the Large Hadron Collider (LHC) at CERN is undergoing upgrades to its instrumentation, as well as the hardware and software that comprise its Trigger and Data Acquisition (TDAQ) system. The increased energy will yield larger cross sections for interesting physics processes, but will also lead to increased artifacts in on-line reconstruction in the trigger, as well as increased trigger rates, beyond the current system’s capabilities. To meet these demands it is likely that the massive parallelism of General-Purpose Programming with Graphic Processing Units (GPGPU) will be utilised. This dissertation addresses the problem of integrating GPGPU into the existing Trigger and TDAQ platforms; detailing and analysing GPGPU performance in the context of performing in a high-throughput, on-line environment like ATLAS. Preliminary tests show low to moderate speed-up with GPU relative to CPU, indicating that to achieve a more significant performance increase it may be necessary to alter the current platform beyond pairing suitable GPUs to CPUs in an optimum ratio. Possible solutions are proposed and recommendations for future work are given.