Introduction
I graduated from the department in 2008 with a Ph.D. in computer science. I have joined Google as a software engineer.
I received my Bachelor's degree in Computer Engineering from the Kandó Kálmán College of Engineering in Budapest, Hungary (currently: Kandó Kálmán Faculty of Electrical Engineering at the Budapest Polytechnic). I received my Master's degree in Computer Science from the Faculty of Science at the University of Szeged in Szeged, Hungary.
Visit my tech blog and my personal site.
Research
My advisors were Prof. Kevin Skadron and Prof. Tarek Abdelzaher (UIUC).
My research area is energy management in real-time server systems. I am especially interested in large-scale Web server farms, where thousands of server machines may be tightly placed in a small space, leading to huge energy consumption and serious thermal problems. Energy management in these systems is becoming crucial. Fortunately, there are many hardware features that can be leveraged: low-power CPU states, dynamic voltage scaling, low-power system states combined with remote (LAN) wakeup features, or even remote-controlled power strips. At the same time, software features such as temperature-aware load balancing can help distribute the heat better, so that cooling is more effective, which is important in order to increase the reliability of hardware components and to minimize the possibility of thermal throttling.
However, the necessary reduction in power consumption also translates to a performance hit. Therefore the power management policy must take into account the real-time performance (delay) constraints of the system. Energy management is especially difficult in multi-tier servers, because the combined delay of all tiers is perceived by the clients as the service delay. As part of my earlier research in this area, I have successfully implemented a prototype three-tier Web server that optimizes energy consumption while providing (soft) real-time performance guarantees.
My latest efforts were focused on generalizing the previous results to multi-tier server clusters, in which each tier consists of multiple machines. The results of this research will hopefully be of great practical value in making the operation of large data centers more economical and environmentally friendly.
Publications
- T. Horvath, T. Abdelzaher, K. Skadron, and X. Liu. "Dynamic Voltage Scaling in Multi-tier Web Servers with End-to-end Delay Control." IEEE Transactions on Computers, 56(4):444-58, Apr. 2007. Earlier version published as TR CS-2004-34, Department of Computer Science, University of Virginia, Nov. 2004.
- T. Horvath, K. Skadron, and T. Abdelzaher. "Enhancing Energy Efficiency in Multi-tier Web Server Clusters via Prioritization." In Proceedings of the 2007 NSF Next Generation Software Workshop, in conjunction with the IEEE International Parallel and Distributed Processing Symposium, Mar. 2007.
- T. Horvath and K. Skadron "Multi-mode Energy Management for Multi-tier Server Clusters." In Proceedings of the ACM/IEEE/IFIP International Conference on Parallel Architectures and Compilation Techniques (PACT), Oct. 2008. [Presentation]
