Lock-Free Parallel Algorithms: An Experimental Study

Abstract

Abstract. Lock-free shared data structures in the setting of distributed computing have received a fair amount of attention. Major motivations of lock-free data structures include increasing fault tolerance of a (possibly heterogeneous) system and getting rid of the problems associated with critical sections such as priority inversion and deadlock. For parallel computers with closely-coupled processors and shared memory, these issues are no longer major concerns. While many of the results are applicable especially when the model used is shared memory multiprocessors, no prior studies have considered improving the performance of a parallel implementation by way of lock-free programming. As a matter of fact, often times in practice lock free data structures in a distributed setting do not perform as well as those that use locks. As the data structures and algorithms for parallel computing are often drastically different from those in distributed computing, it is possible that lock-free programs perform better. In this paper we compare the similarity and difference of lock-free programming in both distributed and parallel computing environments and explore the possibility of adapting lock-free programming to parallel computing to improve performances. Lock-free programming also provides a new way of simulating PRAM and asynchronous PRAM algorithms on current parallel machines.

Software Diversity-Based Active Replication as an Approach for Enhancing the Performance of Advanced Simulation Systems

“…This is in the spirit of our approach, which is oriented to increasing the simulation system execution speed possibly at the expense of a higher amount of resource usage due to the active replication scheme. Additionally, as also recently shown in [5], application level spinlocking reveals an effective alternative to system level blocking approaches while handling critical sections when the target is application level execution speed on SMP systems.…”

A Generic and Highly Efficient Parallel Variant of Boruvka's Algorithm

“…The same authors presented in [11] an algorithmic variant of Borůvka's algorithm that uses colors to denote supervertices, from here on referred to as Cong2005. There are two implementations of this variant, one with platform-specific assembly instructions, which we cannot use for comparisons purposes, and one with pThread mutexes, whose performance is shown in Section V. Our implementation both more generic than Cong2005, since no machine-specific assembly instructions are used, and more efficient for every tested case.…”

“…We thank G. Cong for providing the implementation shown in [11], and R. Nasre and F. Correia for insightful discussions.…”

“…First, we present a parallel and platform independent variant of Borůvka's algorithm that attains high performance and good scalability on multi-core CPU-chips and GPUs, in isolation. Second, we present a comprehensive comparison of the implementations of MST-solvers described in [11], [7], [12] and with the framework described in [13], wherein we include the implementations of our variant, which outperforms all the others.…”

Green-Marl

“…For example, a simple BFS traversal has been implemented differently on commodity servers [4,22], GPUs [21], Cray XMT machines [6] and cluster environments [38]. Techniques for efficient implementation of other algorithms such as betweenness centrality [28], shortest path [27], and minimum spanning tree [16] can be found in the research literature. Green-Marl does not obviate the need for these studies; rather these techniques can easily be reused and applied to other algorithms with similar patterns by means of the Green-Marl DSL and compiler.…”