Efficient multi-word locking using randomization
Phuong Hoai Ha, Philippas Tsigas, Mirjam Wattenhofer, Rogert Wattenhofer · 2005
In this paper we examine the general multi-word lock problem, where processes are allowed to multilock arbitrary registers. Aiming for a highly efficient solution we propose a randomized algorithm which successfully breaks long dependency chains, the crucial factor for slowing down an execution. In the analysis we focus on the 2-word lock problem and show that in this special case an execution of our algorithm takes with high probability at most time O( ∆ 3 log n / log log n), where n is the number of registers and ∆ the maximal number of processes interested in the same register (the contention). Furthermore, we implemented our algorithm for the general multi-word lock problem on an SGI Origin2000 machine, demonstrating that our algorithm is not only of theoretical interest.