Autonomous tools for high performance in large datacentres
Abel Souza has developed algorithms and intelligent software tools to improve the performance of the large computing datacentres that power today’s Internet services and enables scientific simulations. He defends his doctoral thesis on the 8 May at Umeå University.
Text: Ingrid Söderbergh
Abel Souza defends his thesis on the 8th of May at Umeå University.
Datacentres are the main computing infrastructures powering today’s Internet, with many cloud services used for leisure such as Google, Netflix, and Dropbox, as well as work - scientific projects addressing urgent challenges in for example infectious diseases, cancer, and climate change. These infrastructures enable models in science and engineering to be assessed, tested, and evaluated in a timely manner.
However, to provide the world with this amount of computing resources, global datacentres consumed roughly two percent of the electricity worldwide in 2018, with many projections skyrocketing this consumption to ten percent by 2030. From smart sensors used in households to large turbines in remote areas, millions of devices are now connected to cloud datacentres. In this decade we will see datacentres processing yottabyte data amounts (= 1 trillion terabytes), although their computing power will not increase as much due to technological limitations.
These figures show that the demand is soon outpacing the supply, raising serious concerns about datacentre efficiency and slowing down the digitalization of society, with no simple solutions to handle such enormous scales in data.
To tackle these long-term developments, Abel Souza proposes in his thesis autonomic methods combined with novel scheduling strategies to improve datacentre efficiency without compromising in performance. His work develops software architectures and algorithms based on learning, which enable the applications to make the best use possible of high performance datacentre environments.
“An overall theme in my thesis is the extensive experimentation in real datacentres. My results show improvements in datacentre utilization and performance, achieving higher overall efficiency. The methods I have developed also simplify operations and enable the introduction of novel types of applications previously not supported,” says Abel Souza.
There are several trade-offs in allocating datacentre resources according to an application’s workload variations. Understanding these trade-offs give operational efficiency gains because the infrastructure can be tailored to the application needs, instead of the other way around. Autonomic tools to handle this helps developers focus on what matters most to them, while alleviating IT operators managing these complex infrastructures. In summary, a timely allocation of resources to applications ensures infrastructure efficiency, operational cost reductions, and faster time to solutions.
“During my PhD I could experience these challenges first-hand and discuss them with many international researchers. It was very important to my thesis work that the research lab I work in – Autonomous Distributed Systems Lab – has many international connections with both industry and academia,” says Abel Souza.
Abel Souza was born in Brazil and grew up in the state of Rio de Janeiro. He received his Bachelor’s degree in Computer Science from the Federal Fluminense University (UFF) in 2014, with two international experiences in the USA: one year exchange studies at University of Nebraska-Lincoln during his bachelor’s degree, and a 7 months PhD internship at the Lawrence Berkeley National Lab.
About the dissertation:
On Friday May 8th, Abel Souza, the Department of Computing Science at Umeå University, defends his thesis entitled "Autonomous Resource Management for High Performance Datacentres" The dissertation takes place at 10:00 in the MIT Place Seminarierummet, MIT-byggnaden (Plan 2), Umeå University, Sweden Faculty opponent is Professor Dr. Alexandru Iosup, Computer Systems Department, Faculty of Science, Vrije Universiteit Amsterdam, Amsterdam, Netherlands Main Supervisor is Associate Professor Johan Tordsson