article · 2014
Distributed Computing Group
David Hasenfratz, Prof Dr, Roger Wattenhofer, Advisor Johannes Schneider
Cited by 0Abstract and details only
Abstract
Practically any notebook or desktop computer today is equipped with dual-core chips and already quad-core chips are coming to the market. In contrast a huge junk of the available software today is not able to make full use out of these mighty processors. With monitors, semaphores or locks it is hard and time-consuming to write efficient and correct code which scales well with the number of increasing cores. Transactional Memory aims to solve this problem. The idea of Transactional Memory (TM) has been brought from database sys-tems to programming. TM is a hot research topic nowadays, considered by many researchers to be one of the most promising solutions for concurrency control. In a TM system, the contention manager is responsible for resolving the con-flicts between transactions. There are already a number of policies available, how-ever benchmarks have been limited and neither theory, nor the benchmarks consid-ered, yield a clear picture of how a contention manager should make its decision. None of the policies is outperforming all the others on the tested benchmarks. This thesis introduces 10 new contention manager policies. Two of them, the AccurateTimestampBackoffTime and TransactionalArrayReactivateAny man-agers, show very good results. Both outperform all the published managers on three of the tested benchmarks. On the other two benchmarks the TransactionalAr-rayReactivateAny manager performs equally well as the best published policy. On one benchmark the AccurateTimestampBackoffTime policy is worse than the best published manager. The benchmarks were executed on 4 AMD Quadcore Opteron 8350 (2.0 GHz) processors, using the DSTM2.1 Java library from Herlihy et al. Acknowledgment I would like to thank Prof. Dr. Roger Wattenhofer for giving me the opportunity to write my semester thesis at the Distributed Computing Group. Special thanks go to Johannes Schneider who supervised my work very com-petently. Thank you for having an open ear for all my problems and questions at any time during the day. i ii
Cite this paper
Hasenfratz, D., Dr, P., Wattenhofer, R., & Schneider, A. J. (2014). Distributed Computing Group.
Hasenfratz, David, et al. “Distributed Computing Group.” 2014.
Hasenfratz, David, Prof Dr, Roger Wattenhofer, and Advisor Johannes Schneider. 2014. “Distributed Computing Group.”
Hasenfratz, D. et al. (2014) ‘Distributed Computing Group’.
@article{hasenfratz2014distributed,
title = {{Distributed Computing Group}},
author = {Hasenfratz, David and Dr, Prof and Wattenhofer, Roger and Schneider, Advisor Johannes},
year = {2014}
}Made from the index record. Check names, capitalisation and page numbers against the paper. Need another style? AI for Research has over 10,000.