Maximal Antichain Lattice Algorithms for Distributed Computations
Maximal Antichain Lattice Algorithms for Distributed Computations
复制标题
分布式计算的最大反链格算法
DOI:
10.1007/978-3-642-35668-1_17
复制
发表时间:
2013
期刊:
影响因子:
--
通讯作者:
V. Garg
中科院分区:
文献类型:
--
作者:
V. Garg
The lattice of maximal antichains of a distributed computation is generally much smaller than its lattice of consistent global states. We show that a useful class of predicates can be detected on the lattice of maximal antichains instead of the lattice of consistent cuts obtaining significant (exponential for many cases) savings. We then propose new online and offline algorithms to construct and enumerate the lattice of maximal antichains. Previously known algorithm by Nourine and Raynoud [NR99, NR02] to construct the lattice takes O(n 2 m) time where n is the number of events in the computation, and m is the size of the lattice of maximal antichains. The algorithm by Jourdan, Rampon and Jard [JRJ94] takes O((n + w 2)wm) time where w is the width of the computation. All these algorithms assume as input the lattice of maximal antichains prior to the arrival of a new event. We present a new online incremental algorithm, OLMA, that computes the newly added elements to the lattice without requiring the prior lattice. Since the lattice may be exponential in the size of the computation, we get a significant reduction in the space complexity. The OLMA algorithm takes O(mw 2 logw L ) time and O(w L w logn) space where w L is the width of the lattice of maximal antichains. The lower space complexity makes our algorithm applicable for online global predicate detection in a distributed system. For the purposes of analyzing offline traces, we also propose new enumeration algorithms to traverse the lattice.