Sriram V. Pemmaraju

Fault-containing self-stabilizing algorithms

Self-stabilization provides a non-masking approach to fault tolerance. Given this fact, one would hope that in a self-stabilizing system, the amount of disruption caused by a fault is proportional to the severity of the fault. However, this is not true for many self-stabilizing systems. Our paper addresses this weakness of distributed self-stabilizing systems by introducing the notion of fault containment. Informally, a fault-containing self-stabilizing algorithm is one that contains the effects of limited transient faults while retaining the property of self-st abilization. The paper begins with a formal framework for specifying and evaluating fault-containing self-stabilizing protocols. Then, it is shown that self-stabilization and fault containment are goals that can conflict. For example, it is shown that imposing a O(1) bound on the worst case recovery time from a l-faulty state necessitates added overhead for stabilization: for some tasks, the O(1) recovery time implies sfiabilization time cannot be within O(1) rounds from the optimum value. The paper then presents a transformer T that maps any non-reactive self-stabilizing algorithm P into an equivalent fault-containing self-stabilizing algorithm Pf that can repair any l-faulty state in O(1) time with O(1) space overhead. This transformation is baaed on a novel stabilizing timer paradigm that significantly simplifies the ti=k of fault containment. The paper concludes by generalizing the transformer ‘T into a parameterized transformer 7(k) such that for varying k we obtain varying performance measures for Pf.

Stack and Queue Layouts of Directed Acyclic Graphs: Part I

Stack layouts and queue layouts of undirected graphs have been used to model problems in fault-tolerant computing and in parallel process scheduling. However, problems in parallel process scheduling are more accurately modeled by stack and queue layouts of directed acyclic graphs (dags). A stack layout of a dag is similar to a stack layout of an undirected graph, with the additional requirement that the nodes of the dag be in some topological order. A queue layout is defined in an analogous manner. The stacknumber ( queuenumber) of a dag is the smallest number of stacks (queues) required for its stack layout (queue layout). In this paper, bounds are established on the stacknumber and queuenumber of two classes of dags: tree dags and unicyclic dags. In particular, any tree dag can be laid out in 1 stack and in at most 2 queues; and any unicyclic dag can be laid out in at most 2 stacks and in at most 2 queues. Forbidden subgraph characterizations of 1-queue tree dags and 1-queue cycle dags are also presented. Part II of this paper presents algorithmic results---in particular, linear time algorithms for recognizing 1-stack dags and 1-queue dags and proof of NP-completeness for the problem of recognizing a 4-queue dag and the problem of recognizing a 9-stack dag.

Greedy Routing with Bounded Stretch

Greedy routing is a novel routing paradigm where messages are always forwarded to the neighbor that is closest to the destination. Our main result is a polynomial-time algorithm that embeds combinatorial unit disk graphs (CUDGs - a CUDG is a UDG without any geometric information) into O(log 2 n)- dimensional space, permitting greedy routing with constant stretch. To the best of our knowledge, this is the first greedy embedding with stretch guarantees for this class of networks. Our main technical contribution involves extracting, in polynomial time, a constant number of isometric and balanced tree separators from a given CUDG. We do this by extending the celebrated Lipton-Tarjan separator theorem for planar graphs to CUDGs. Our techniques extend to other classes of graphs; for example, for general graphs, we obtain an O(log n)-stretch greedy embedding into O(log 2 n)-dimensional space. The greedy embeddings constructed by our algorithm can also be viewed as a constant-stretch compact routing scheme in which each node is assigned an O(log 3 n)-bit label. To the best of our knowledge, this result yields the best known stretch-space trade-off for compact routing on CUDGs. Extensive simulations on random wireless networks indicate that the average routing overhead is about 10%; only few routes have a stretch above 1.5.

Energy conservation via domatic partitions

Using a dominating set as a coordinator in wireless networks has been proposed in many papers as an energy conservation technique. Since the nodes in a dominating set have the extra burden of coordination, energy resources in such nodes will drain out more quickly than in other nodes. To maximize the lifetime of nodes in the network,it has been proposed that the role of coordinators be rotated among the nodes in the network. One abstraction that has been considered for the problem of picking a collection of coordinators and cycling through them, is the domatic partition problem. This is the problem of partitioning the set of the nodes of the network into dominating sets with the aim of maximizing the number of dominating sets. In this paper,we consider the k -domatic partition problem. A k -dominating set is a subset D of nodes such that every node in the network is at distance at most k from D. The k-domatic partition problem seeks to partition the network into maximum number of k-dominating sets.We point out that from the point of view of saving energy,it may be better to construct a k-domatic partition for k >1.We present three deterministic, distributed algorithms for finding large k-domatic partitions for k > 1. Each of our algorithms constructs a k-domatic partition of size at least a constant fraction of the largest possible (k 1)-domatic partition. Our first algorithm runs in constant time on unit ball graphs (UBGs) in Euclidean space assuming that all nodes know their positions in a global coordinate system. Our second algorithm drops knowledge of global coordinates and instead assumes that pairwise distances between neighboring nodes are known. This algorithm runs in O(log* n ) time on UBGs in a metric space with constant doubling dimension. Our third algorithm drops all reliance on geometric information, using connectivity information only. This algorithm runs in O(log Δ · log *n) time on growth-bounded graphs. Euclidean UBGs, UBGs in metric spaces with constant doubling dimension, and growth-bounded graphs are successively more general models of wireless networks and all three models include the well-known, but somewhat simplistic wireless network models such as unit disk graphs.

Approximation algorithms for the max-coloring problem

Given a graph G = (V, E) and positive integral vertex weights w : V → N, the max-coloring problem seeks to find a proper vertex coloring of G whose color classes C1, C2, ..., Ck, minimize

Self-Stabilizing Algorithms for Finding Centers and Medians of Trees

{\sum_{i=1}^{k}}{\it max}_{v\epsilon C{_{i}} {\it w}(v)}

Stack and Queue Layouts of Posets

. The problem arises in scheduling conflicting jobs in batches and in minimizing buffer size in dedicated memory managers. In this paper we present three approximation algorithms and one inapproximability result for the max-coloring problem. We show that if for a class of graphs

New Results for the Minimum Weight Triangulation Problem

{\mathcal G}

Error-detecting codes and fault-containing self-stabilization

, the classical problem of finding a proper vertex coloring with fewest colors has a c-approximation, then for that class

Recognizing Leveled-Planar Dags in Linear Time

{\mathcal G}