In this work, we consider contention management in transactional memory in the context of balanced workloads which exhibit better performance potential. Balanced workloads include transactions whose number of exclusive accesses to the shared resources for writes are within a constant fraction of the total accesses (read or writes) to the shared resources. We explore the theoretical performance boundaries of contention management for balanced workloads from the worst-case perspective. We present and analyze two new randomized contention management algorithms which achieve competitive ratio very close to O( √ s), where s is the number of shared resources. We also prove a matching lower bound in competitive ratio by showing that no contention management algorithm can be better than O( √ s)competitive ratio for balanced workloads, unless NP⊆ZPP. To our knowledge, these results demonstrate a significant improvement over the previously best bound of O(s).
[1]
Subhash Khot,et al.
Improved inapproximability results for MaxClique, chromatic number and approximate graph coloring
,
2001,
Proceedings 2001 IEEE International Conference on Cluster Computing.
[2]
Rachid Guerraoui,et al.
Toward a theory of transactional contention managers
,
2005,
PODC '05.
[3]
Hagit Attiya,et al.
Transactional Contention Management as a Non-Clairvoyant Scheduling Problem
,
2010,
Algorithmica.
[4]
Roger Wattenhofer,et al.
Bounds on Contention Management Algorithms
,
2009,
ISAAC.
[5]
Hagit Attiya,et al.
Transactional scheduling for read-dominated workloads
,
2009,
J. Parallel Distributed Comput..