arXiv subjects

Sriram V. Pemmaraju

Publications and source records attributed to Sriram V. Pemmaraju.

24 records · Page 2Linked to original sources

arXiv2014

Minimum-weight Spanning Tree Construction in $O(\log \log \log n)$ Rounds on the Congested Clique

This paper considers the \textit{minimum spanning tree (MST)} problem in the Congested Clique model and presents an algorithm that runs in $O(\log \log \log n)$ rounds, with high probability. Prior to this, the fastest MST algorithm in this model was a deterministic algorithm due to Lotker et al.~(SIAM J on Comp, 2005) from about a decade ago. A key step along the way to designing this MST algorithm is a \textit{connectivity verification} algorithm that not only runs in $O(\log \log \log n)$ rounds with high probability, but also has low message complexity. This allows the fast computation of an MST by running multiple instances of the connectivity verification algorithm in parallel. These results depend on a new edge-sampling theorem, developed in the paper, that says that if each edge $e = \{u, v\}$ is sampled independently with probability $c \log^2 n/\min\{\mbox{degree}(u), \mbox{degree}(v)\}$ (for a large enough constant $c$) then all cuts of size at least $n$ are approximated in the sampled graph. This sampling theorem is inspired by series of papers on graph sparsification via random edge sampling due to Karger~(STOC 1994), Benczúr and Karger~(STOC 1996, arxiv 2002), and Fung et al.~(STOC 2011). The edge sampling techniques in these papers use probabilities that are functions of edge-connectivity or a related measure called edge-strength. For the purposes of this paper, these edge-connectivity measures seem too costly to compute and the main technical contribution of this paper is to show that degree-based edge-sampling suffices to approximate large cuts.

cs.DC↗

arXiv2014

Lessons from the Congested Clique Applied to MapReduce

The main results of this paper are (I) a simulation algorithm which, under quite general constraints, transforms algorithms running on the Congested Clique into algorithms running in the MapReduce model, and (II) a distributed $O(Δ)$-coloring algorithm running on the Congested Clique which has an expected running time of (i) $O(1)$ rounds, if $Δ\geq Θ(\log^4 n)$; and (ii) $O(\log \log n)$ rounds otherwise. Applying the simulation theorem to the Congested-Clique $O(Δ)$-coloring algorithm yields an $O(1)$-round $O(Δ)$-coloring algorithm in the MapReduce model. Our simulation algorithm illustrates a natural correspondence between per-node bandwidth in the Congested Clique model and memory per machine in the MapReduce model. In the Congested Clique (and more generally, any network in the $\mathcal{CONGEST}$ model), the major impediment to constructing fast algorithms is the $O(\log n)$ restriction on message sizes. Similarly, in the MapReduce model, the combined restrictions on memory per machine and total system memory have a dominant effect on algorithm design. In showing a fairly general simulation algorithm, we highlight the similarities and differences between these models.

cs.DC↗

arXiv2013

Super-Fast Distributed Algorithms for Metric Facility Location

This paper presents a distributed O(1)-approximation algorithm, with expected-$O(\log \log n)$ running time, in the $\mathcal{CONGEST}$ model for the metric facility location problem on a size-$n$ clique network. Though metric facility location has been considered by a number of researchers in low-diameter settings, this is the first sub-logarithmic-round algorithm for the problem that yields an O(1)-approximation in the setting of non-uniform facility opening costs. In order to obtain this result, our paper makes three main technical contributions. First, we show a new lower bound for metric facility location, extending the lower bound of Bădoiu et al. (ICALP 2005) that applies only to the special case of uniform facility opening costs. Next, we demonstrate a reduction of the distributed metric facility location problem to the problem of computing an O(1)-ruling set of an appropriate spanning subgraph. Finally, we present a sub-logarithmic-round (in expectation) algorithm for computing a 2-ruling set in a spanning subgraph of a clique. Our algorithm accomplishes this by using a combination of randomized and deterministic sparsification.

cs.DC↗

arXiv2013

A Super-Fast Distributed Algorithm for Bipartite Metric Facility Location

The \textit{facility location} problem consists of a set of \textit{facilities} $\mathcal{F}$, a set of \textit{clients} $\mathcal{C}$, an \textit{opening cost} $f_i$ associated with each facility $x_i$, and a \textit{connection cost} $D(x_i,y_j)$ between each facility $x_i$ and client $y_j$. The goal is to find a subset of facilities to \textit{open}, and to connect each client to an open facility, so as to minimize the total facility opening costs plus connection costs. This paper presents the first expected-sub-logarithmic-round distributed O(1)-approximation algorithm in the $\mathcal{CONGEST}$ model for the \textit{metric} facility location problem on the complete bipartite network with parts $\mathcal{F}$ and $\mathcal{C}$. Our algorithm has an expected running time of $O((\log \log n)^3)$ rounds, where $n = |\mathcal{F}| + |\mathcal{C}|$. This result can be viewed as a continuation of our recent work (ICALP 2012) in which we presented the first sub-logarithmic-round distributed O(1)-approximation algorithm for metric facility location on a \textit{clique} network. The bipartite setting presents several new challenges not present in the problem on a clique network. We present two new techniques to overcome these challenges. (i) In order to deal with the problem of not being able to choose appropriate probabilities (due to lack of adequate knowledge), we design an algorithm that performs a random walk over a probability space and analyze the progress our algorithm makes as the random walk proceeds. (ii) In order to deal with a problem of quickly disseminating a collection of messages, possibly containing many duplicates, over the bipartite network, we design a probabilistic hashing scheme that delivers all of the messages in expected-$O(\log \log n)$ rounds.

cs.DC↗

arXiv2012

On the Analysis of a Label Propagation Algorithm for Community Detection

This paper initiates formal analysis of a simple, distributed algorithm for community detection on networks. We analyze an algorithm that we call \textsc{Max-LPA}, both in terms of its convergence time and in terms of the "quality" of the communities detected. \textsc{Max-LPA} is an instance of a class of community detection algorithms called \textit{label propagation} algorithms. As far as we know, most analysis of label propagation algorithms thus far has been empirical in nature and in this paper we seek a theoretical understanding of label propagation algorithms. In our main result, we define a clustered version of \er random graphs with clusters $V_1, V_2,..., V_k$ where the probability $p$, of an edge connecting nodes within a cluster $V_i$ is higher than $p'$, the probability of an edge connecting nodes in distinct clusters. We show that even with fairly general restrictions on $p$ and $p'$ ($p = Ω(\frac{1}{n^{1/4-ε}})$ for any $ε> 0$, $p' = O(p^2)$, where $n$ is the number of nodes), \textsc{Max-LPA} detects the clusters $V_1, V_2,..., V_n$ in just two rounds. Based on this and on empirical results, we conjecture that \textsc{Max-LPA} can correctly and quickly identify communities on clustered \er graphs even when the clusters are much sparser, i.e., with $p = \frac{c\log n}{n}$ for some $c > 1$.

cs.DC↗

arXiv2008

Localized Spanners for Wireless Networks

We present a new efficient localized algorithm to construct, for any given quasi-unit disk graph G=(V,E) and any e > 0, a (1+e)-spanner for G of maximum degree O(1) and total weight O(w(MST)), where w(MST) denotes the weight of a minimum spanning tree for V. We further show that similar localized techniques can be used to construct, for a given unit disk graph G = (V, E), a planar Cdel(1+e)(1+pi/2)-spanner for G of maximum degree O(1) and total weight O(w(MST)). Here Cdel denotes the stretch factor of the unit Delaunay triangulation for V. Both constructions can be completed in O(1) communication rounds, and require each node to know its own coordinates.

cs.DC↗