2026-10-13T09:50:29Z https://api.figshare.com/v2/oai

oai:figshare.com:article/32805680 2026-10-01T16:14:10Z portal_1172 item_type_8 month_year_10_2026

Improving run-time performance of server-based data processing applications Jordan Christain Matthew Leggett (24292859) PUREID: 685727105 Data Centre FPGA scheduling The continuous growth of data centres keeps pushing the demand for better resource utilisation in order to retain efficiency. Whilst hardware capabilities of data centres keeps improving, the software running needs to be able to make use of these new capabilities and meet emerging requirements. This thesis presents research for improving the performance of data centres by making use of the specialised hardware and introducing a software framework to allow the performance to be analysed.<br><br>The first optimisation presented optimises a specific function in some database systems called the delta merge function by providing an field programmable gate array based solution. Integration in a production database has shown a speedup over 100x for certain cases compared to a central processing unit implementation, and 2x in the worst case.<br><br>In addition to the acceleration of databases, this research focuses on improving the performance of general tasks in data centres by optimising the scheduler. Dynamic binary instrumentation tools allow a program to be intercepted and modified in runtime, but operate at a low level view. To address this, the PinCFG framework was created to allow an application control flow graph can be created, allowing the user to analyse the control flow dynamically. The overhead was compared to running on Pin alone without the Control Flow Graph (CFG), and had mixed results. In all cases the CFG construction added 10%, and in the worst case over 8000% overhead was introduced. Simulations were ran and a proposal to improve the PinIT tool by adding dynamic CFG generation is presented. 2026-10-01T16:14:10Z Text Thesis 10.17034/32805680.v1 https://figshare.com/articles/thesis/Improving_run-time_performance_of_server-based_data_processing_applications/32805680 All Rights Reserved