Punit Chandra

Causal multicast in mobile networks

Distributed applications using file sharing and file replication, conferencing, and delivery in multimedia systems often need to use the semantics of causal multicast. Causal multicast in a mobile system is complicated by the various limitations imposed by the mobile system. Though there are different directions to improve the performance of causal multicast algorithms in mobile systems, at the core of such algorithms is the module to impose causal delivery among a set of nodes, which may be either mobile hosts, or mobile service stations, or peers at a higher level in the interconnection hierarchy. We show how to adapt the optimal causal multicast algorithm of Kshemkalyani-Singhal to mobile networks, and then show simulation results comparing its performance with that of other algorithms for causal multicast in mobile networks.

Performance of the optimal causal multicast algorithm: a statistical analysis

An optimal causal message ordering algorithm for asynchronous distributed systems was proposed by Kshemkalyani and Singhal and its optimality was proven theoretically. For a system of n processes, although the space complexity of this algorithm was shown to be O(n/sup 2/) integers, it was expected that the actual space overhead would be much less than n/sup 2/. It is difficult to determine the behavior of this algorithm by a theoretical analysis. We measure the overheads of two different implementations of the optimal causal message ordering algorithm via simulation under a wide range of system conditions. The optimal algorithm is seen to display significantly less message space overhead and log space overhead than the canonical Raynal-Schiper-Toueg algorithm.

Distributed algorithm to detect strong conjunctive predicates

This paper presents an on-line distributed algorithm for detection of Definitely(φ) for the class of conjunctive global predicates. The only known algorithm for detection of Definitely(φ) uses a centralized approach. A method for decentralizing the algorithm was also given, but the work load is not fairly distributed and the method uses a hierarchical structure. The centralized approach has a time, space, and total message complexity of O(n2m), where n is the number of processes and m is the maximum number of messages sent by any process. The proposed on-line distributed algorithm uses the concept of intervals rather than events, and assumes p is the maximum number of intervals at any process. The worst-case time complexity across all the processes is O(min(pn2, mn2)). The worst-case space overhead across all the processes is min(2mn2, 2pn2).

Detection of Orthogonal Interval Relations

The complete set R of orthogonal temporal interactions between pairs of intervals, formulated by Kshemkalyani, allows the detailed specification of the manner in which intervals can be related to one another in a distributed execution. This paper presents a distributed algorithm to detect whether pre-specified interaction types between intervals at different processes hold. Specifically, for each pair of processes i and j, given a relation ri, j from the set of orthogonal relations R, this paper presents a distributed (on-line) algorithm to determine the intervals, if they exist, one from each process, such that each relation ri, j is satisfied for that (i, j) process pair. The algorithm uses O(n min(np, 4mn)) messages of size O(n) each, where n is the number of processes, m is the maximum number of messages sent by any process, and p is the maximum number of intervals at any process. The average time complexity per process is O(min(np, 4mn)), and the total space complexity across all the processes is min(4pn2 - 2np, 10mn2).

Global state detection based on peer-to-peer interactions

This paper presents an algorithm for global state detection based on peer-to-peer interactions. The interactions in distributed systems can be analyzed in terms of the peer-to-peer pairwise interactions of intervals between processes. This paper examines the problem: “If a global state of interest to an application is specified in terms of the pairwise interaction types between each pair of peer processes, how can such a global state be detected?” Devising an efficient algorithm is a challenge because of the overhead of having to track the intervals at different processes. We devise a distributed on-line algorithm to efficiently manage the distributed data structures and solve this problem. We prove the correctness of the algorithm and analyze its complexity.

Global Predicate Detection under Fine-Grained Modalities

Predicate detection is an important problem in distributed systems. Based on the temporal interactions of intervals, there exists a rich class of modalities under which global predicates can be specified. For a conjunctive predicate φ, we show how to detect the traditional Possibly(φ) and Definitely(φ) modalities along with the added information of the exact interaction type between each pair of intervals (one interval at each process). The polynomial time, space, and message complexities of the proposed on-line detection algorithms to detect Possibly andDefinitely in terms of the fine-grained interaction types per pair of processes, are the same as those of the earlier on-line algorithms that can detect only whether thePossibly andDefinitely modalities hold.

Analysis of interval-based global state detection

The problem of global state observation is fundamental to distributed systems. All interactions in distributed systems can be analyzed in terms of the building block formed by the pairwise interactions of intervals between two processes. Considering causality-based pairwise interactions by which two intervals at different processes may interact with each other, there are 40 possible orthogonal interactions. This paper examines the problem: “If a global state of interest to an application is specified in terms of the pairwise interaction types between each pair of processes, how can such a global state be detected?” A solution identifies a global state in which the relation specified for each process pair is satisfied. This paper formulates the specific conditions on the exact communication structures to determine which of the intervals being examined at any time may never satisfy the stipulated relation for that pair of processes, and therefore that interval must be deleted.

Compact Routing in Directed Networks with Stretch Factor of Two

This paper presents a compact routing algorithm with stretch less than 3 for directed networks. Although for stretch less than 3, the lower bound for the total routing information in the network is ?(n2) bits, it is still worth examining to determine the best possible saving in space. The routing algorithm uses header size of 4 log n and provides round-trip stretch factor of 2, while bounding the local space by [n - √n(?1 - 7/4n) - 1/2] log n. These results for stretch less than 3 for directed networks match those for undirected networks.

Efficient Synchronization of Asynchronous Processes

Concurrent programming languages including CSP and Ada use synchronous message-passing to define communication between a pair ofa synchronous processes. This paper presents an efficient way to synchronize these processes by improving on Bagrodias algorithm that provides binary rendezvous. Simulation results are presented to show the better performance of the optimized algorithm for two cases - the case where the interaction set is composed ofall possible pairs and the case where the set ofn ext allowable interactions is ofcard inality one. For the latter, the optimized algorithm also improves upon the best case delay for synchronization. The client-server computing model, the producer-consumer interaction, and interaction between processes executing parallelized tasks represent some broad classes of computations which can leverage the proposed improvements.