## Distributed Approaches to Triangulation and Embedding (2006)

### Cached

### Download Links

- [research.microsoft.com]
- [research.microsoft.com]
- [research.microsoft.com]
- [research.microsoft.com]
- DBLP

### Other Repositories/Bibliography

Citations: | 30 - 7 self |

### BibTeX

@MISC{Slivkins06distributedapproaches,

author = {Aleksandrs Slivkins},

title = {Distributed Approaches to Triangulation and Embedding},

year = {2006}

}

### Years of Citing Articles

### OpenURL

### Abstract

A number of recent papers in the networking community study the distance matrix defined by the node-to-node latencies in the Internet and, in particular, provide a number of quite successful distributed approaches that embed this distance into a low-dimensional Euclidean space. In such algorithms it is feasible to measure distances among only a linear or near-linear number of node pairs; the rest of the distances are simply not available. Moreover, for applications it is desirable to spread the load evenly among the participating nodes. Indeed, several recent studies use this ’fully distributed ’ approach and achieve, empirically, a low distortion for all but a small fraction of node pairs. This is concurrent with the large body of theoretical work on metric embeddings, but there is a fundamental distinction: in the theoretical approaches to metric embeddings, full and centralized access to the distance matrix is assumed and heavily used. In this paper we present the first fully distributed embedding algorithm with provable distortion guarantees for doubling metrics (which have been proposed as a reasonable abstraction of Internet latencies), thus providing some insight into the empirical success of the recent Vivaldi algorithm [7]. The main ingredient of our embedding algorithm is an improved fully distributed algorithm for a more basic problem of triangulation, where the triangle inequality is used to infer the distances that have not been measured; this problem received a considerable attention in the networking community, and has also been studied theoretically in [19]. We use our techniques to extend ɛ-relaxed embeddings and triangulations to infinite metrics and arbitrary measures, and to improve on the approximate distance labeling scheme of Talwar [36].

### Citations

1923 | Randomized Algorithms - Motwani, Raghavan - 1995 |

557 | Predicting Internet Network Distance with Coordinates-based Approaches
- Ng, Zhang
- 2002
(Show Context)
Citation Context ...e latencies in the Internet 1 (e.g. [9, 13, 8, 20, 37]) and, in particular, provide a number of quite successful distributed approaches that embed this distance into a low-dimensional Euclidean space =-=[30, 7, 32, 5, 25]-=-. In such algorithms it is feasible to measure distances among only a linear or near-linear number of node pairs; the rest of the distances are simply not available. For instance, the Global Network P... |

517 | Accessing nearby copies of replicated objects in a distributed environment
- Plaxton, Rajaraman, et al.
- 1999
(Show Context)
Citation Context ...6, 23, 22, 19, 28, 4]. At the same time, several recent studies suggest the bounded growth rate of balls as a useful way to capture the structural properties of the Internet distance matrix (see e.g. =-=[34, 8, 30, 31, 41, 15]-=-). Our contributions: In this paper we settle a question left open by [19]: we show that there exists a fully distributed algorithm that embeds a doubling metric into a low-dimensional Euclidean space... |

503 | Vivaldi: A Decentralized Network Coordinate System
- Dabek, Cox, et al.
- 2004
(Show Context)
Citation Context ...on guarantees for doubling metrics (which have been proposed as a reasonable abstraction of Internet latencies), thus providing some insight into the empirical success of the recent Vivaldi algorithm =-=[7]-=-. The main ingredient of our embedding algorithm is an improved fully distributed algorithm for a more basic problem of triangulation, where the triangle inequality is used to infer the distances that... |

501 | Tapestry: A resilient global-scale overlay for service deployment
- Zhao, Huang, et al.
- 2004
(Show Context)
Citation Context ...6, 23, 22, 19, 28, 4]. At the same time, several recent studies suggest the bounded growth rate of balls as a useful way to capture the structural properties of the Internet distance matrix (see e.g. =-=[34, 8, 30, 31, 41, 15]-=-). Our contributions: In this paper we settle a question left open by [19]: we show that there exists a fully distributed algorithm that embeds a doubling metric into a low-dimensional Euclidean space... |

455 | The Geometry of Graphs and some of Its Algorithmic Applications
- Linial, London, et al.
- 1995
(Show Context)
Citation Context ...theory of algorithmic metric embeddings (e.g. [18]) where a full access to the distance matrix is assumed and heavily used. In particular, in the embedding algorithm of Bourgain [3] and Linial et al. =-=[26]-=-, the coordinates are formed by measuring the distance from a point to a set; these sets can be as large as a constant fraction of nodes in a way that would not be feasible to implement for all nodes ... |

278 |
On Lipschitz embedding of finite metric spaces in Hilbert space
- Bourgain
- 1985
(Show Context)
Citation Context ... by the existing rich theory of algorithmic metric embeddings (e.g. [18]) where a full access to the distance matrix is assumed and heavily used. In particular, in the embedding algorithm of Bourgain =-=[3]-=- and Linial et al. [26], the coordinates are formed by measuring the distance from a point to a set; these sets can be as large as a constant fraction of nodes in a way that would not be feasible to i... |

269 | IDMaps: a global Internet host distance estimation service, in
- Francis, Jamin, et al.
- 2000
(Show Context)
Citation Context ...te distance labeling scheme of Talwar [36]. 1 Introduction A number of recent papers in the networking community study the distance matrix defined by the node-tonode latencies in the Internet 1 (e.g. =-=[9, 13, 8, 20, 37]-=-) and, in particular, provide a number of quite successful distributed approaches that embed this distance into a low-dimensional Euclidean space [30, 7, 32, 5, 25]. In such algorithms it is feasible ... |

209 | The Probabilistic Method. WileyInterscience Series - Alon, Spencer - 1992 |

194 |
Distance Geometry and Molecular Conformation
- Crippen, Havel
- 1988
(Show Context)
Citation Context ...tudy here, and makes use of different sampling models and objective functions. Finally, our use of triangulation corresponds to the notion of triangle inequality bounds smoothing in distance geometry =-=[6]-=-, but our setting is different. Preliminaries Denote the uv-distance by duv. Let n be the number of nodes. Call a set of node pairs a ɛ-dense if for each node u it contains at least (1 − ɛ)n pairs uv.... |

188 | Random walks in peer-to-peer networks - Gkantsidis, Mihail, et al. - 2004 |

185 | Distributed Object Location in a Dynamic Network
- Hildrum, Kubiatowicz, et al.
- 2002
(Show Context)
Citation Context ... be seen as an overlay network, our construction is similar in spirit to the overlay topologies constructed for locality-aware distributed hash tables and distributed nearest neighbor selection (e.g. =-=[34, 16, 15, 41]-=-), most notably to the topology constructed in [39]. However, our construction is quite different on the technical level since it is designed to yield provable guarantees on doubling metrics, and is t... |

165 | PIC: Practical Internet Coordinates for Distance Estimation
- Costa, Castro, et al.
- 2004
(Show Context)
Citation Context ...e latencies in the Internet 1 (e.g. [9, 13, 8, 20, 37]) and, in particular, provide a number of quite successful distributed approaches that embed this distance into a low-dimensional Euclidean space =-=[30, 7, 32, 5, 25]-=-. In such algorithms it is feasible to measure distances among only a linear or near-linear number of node pairs; the rest of the distances are simply not available. For instance, the Global Network P... |

156 | Bounded geometry, fractals, and low-distortion embeddings
- Gupta, Krauthgamer, et al.
- 2003
(Show Context)
Citation Context ...g as the doubling dimension of the metric is small. Here the doubling dimension of a metric is defined as the smallest k such that every ball can be covered by 2 k other balls of half the radius (see =-=[2, 12]-=-); a metric with this property is also called 2 k -doubling or just doubling if k is a constant. Doubling metrics, which generalize the distance matrices of low-dimensional point sets in Lp, have been... |

152 | Locating Nearby Copies of Replicated Internet Servers
- Guyton, Schwartz
- 1995
(Show Context)
Citation Context ...te distance labeling scheme of Talwar [36]. 1 Introduction A number of recent papers in the networking community study the distance matrix defined by the node-tonode latencies in the Internet 1 (e.g. =-=[9, 13, 8, 20, 37]-=-) and, in particular, provide a number of quite successful distributed approaches that embed this distance into a low-dimensional Euclidean space [30, 7, 32, 5, 25]. In such algorithms it is feasible ... |

148 | Lighthouses for scalable distributed location
- Pias, Crowcroft, et al.
- 2003
(Show Context)
Citation Context ...e latencies in the Internet 1 (e.g. [9, 13, 8, 20, 37]) and, in particular, provide a number of quite successful distributed approaches that embed this distance into a low-dimensional Euclidean space =-=[30, 7, 32, 5, 25]-=-. In such algorithms it is feasible to measure distances among only a linear or near-linear number of node pairs; the rest of the distances are simply not available. For instance, the Global Network P... |

148 | Meridian: A lightweight network location service without virtual coordinates
- Wong, Slivkins, et al.
- 2005
(Show Context)
Citation Context ...spirit to the overlay topologies constructed for locality-aware distributed hash tables and distributed nearest neighbor selection (e.g. [34, 16, 15, 41]), most notably to the topology constructed in =-=[39]-=-. However, our construction is quite different on the technical level since it is designed to yield provable guarantees on doubling metrics, and is tailored to the specific problem of triangulation. A... |

132 |
On spreading a rumor
- Pittel
- 1987
(Show Context)
Citation Context ... δ = c/ log k and compute an (ɛ, δ)-triangulation using the algorithm from Section 2. Then the beacons measure distances to one another and broadcast them to the entire network using a uniform gossip =-=[33]-=-; in this phase each beacon broadcasts one message of size O(k), the total per-node load being at most O(k 2 log n). Upon receiving this information nodes update the bounds D + on their distances to b... |

127 | Navigating nets: simple algorithms for proximity search
- Krauthgamer, Lee
- 2004
(Show Context)
Citation Context ...ubling metrics, which generalize the distance matrices of low-dimensional point sets in Lp, have been studied recently in the context of metric embeddings, nearest neighbour search and other problems =-=[12, 24, 36, 23, 22, 19, 28, 4]-=-. At the same time, several recent studies suggest the bounded growth rate of balls as a useful way to capture the structural properties of the Internet distance matrix (see e.g. [34, 8, 30, 31, 41, 1... |

118 | Algorithmic applications of low-distortion geometric embeddings
- Indyk
- 2001
(Show Context)
Citation Context ...ointed out by Kleinberg et al. [19], the above Internet-related setting gives rise to a new set of theoretical questions not covered by the existing rich theory of algorithmic metric embeddings (e.g. =-=[18]-=-) where a full access to the distance matrix is assumed and heavily used. In particular, in the embedding algorithm of Bourgain [3] and Linial et al. [26], the coordinates are formed by measuring the ... |

111 |
Lectures on analysis on metric spaces, Universitext
- Heinonen
- 2001
(Show Context)
Citation Context ...(log 2 n). This leads to a (1 + δ)-approximate distance labeling scheme with O(log 2 n log log ∆) bits per label, where ∆ is the aspect ratio of the metric. we will use the notion of doubling measure =-=[14]-=-. 13sTheorem 6.1. For any δ>0 any s-doubling metric has a (0,δ)-triangulation of order O( 1 δ )2 log s log 2 n. Moreover, such triangulation can be computed by an efficient centralized algorithm. Proo... |

102 | Distance labeling in graphs
- Gavoille, Peleg, et al.
(Show Context)
Citation Context ...on. Our contributions: we improve the per-node load for a strong (ɛ, δ)-triangulation to f(ɛ,δ,s)(log n) 4 . Distance labeling We extend our techniques to obtain approximate distance labeling schemes =-=[10]-=- for doubling metrics. Specifically, for any fixed δ > 0 we obtain a (0,δ)-triangulation of order O(log 2 n) (which is of independent interest) and convert it to a (1 + δ)-approximate distance labelin... |

101 | Fast construction of nets in low dimensional metrics, and their applications
- Har-Peled, Mendel
(Show Context)
Citation Context ...ubling metrics, which generalize the distance matrices of low-dimensional point sets in Lp, have been studied recently in the context of metric embeddings, nearest neighbour search and other problems =-=[12, 24, 36, 23, 22, 19, 28, 4]-=-. At the same time, several recent studies suggest the bounded growth rate of balls as a useful way to capture the structural properties of the Internet distance matrix (see e.g. [34, 8, 30, 31, 41, 1... |

84 | Measured descent: A new embedding method for finite metrics
- Krauthgamer, Lee, et al.
(Show Context)
Citation Context ...ubling metrics, which generalize the distance matrices of low-dimensional point sets in Lp, have been studied recently in the context of metric embeddings, nearest neighbour search and other problems =-=[12, 24, 36, 23, 22, 19, 28, 4]-=-. At the same time, several recent studies suggest the bounded growth rate of balls as a useful way to capture the structural properties of the Internet distance matrix (see e.g. [34, 8, 30, 31, 41, 1... |

79 |
Plongements lipschitziens dans
- Assouad
- 1983
(Show Context)
Citation Context ...g as the doubling dimension of the metric is small. Here the doubling dimension of a metric is defined as the smallest k such that every ball can be covered by 2 k other balls of half the radius (see =-=[2, 12]-=-); a metric with this property is also called 2 k -doubling or just doubling if k is a constant. Doubling metrics, which generalize the distance matrices of low-dimensional point sets in Lp, have been... |

76 |
Triangulation and Embedding using Small Sets of Beacons
- Slivkins, Kleinberg, et al.
- 2004
(Show Context)
Citation Context ... triangle inequality is used to infer the distances that have not been measured; this problem received a considerable attention in the networking community, and has also been studied theoretically in =-=[19]-=-. We use our techniques to extend ɛ-relaxed embeddings and triangulations to infinite metrics and arbitrary measures, and to improve on the approximate distance labeling scheme of Talwar [36]. 1 Intro... |

61 | Using Random Subsets to build Scalable Network Services. USITS 03
- Kostic, Rodriguez, et al.
- 2003
(Show Context)
Citation Context ... communicate with (and, in particular, measure distance to) any other node given its ID. Secondly, every node can select a node ID independently and uniformly at random among all nodes in the network =-=[21, 38]-=-. Such operation induces load on multiple nodes; to account for it, let us assume that when each node selects one random node ID, this induces a per-node load of (log n). We call such networks uar-add... |

59 |
Large-scale topological and dynamical properties of the Internet
- Vazquez, Pastor-Satorras, et al.
- 2002
(Show Context)
Citation Context ...te distance labeling scheme of Talwar [36]. 1 Introduction A number of recent papers in the networking community study the distance matrix defined by the node-tonode latencies in the Internet 1 (e.g. =-=[9, 13, 8, 20, 37]-=-) and, in particular, provide a number of quite successful distributed approaches that embed this distance into a low-dimensional Euclidean space [30, 7, 32, 5, 25]. In such algorithms it is feasible ... |

44 |
Routing information organization to support scalable interdomain routing with heterogeneous path requirements
- Hotz
- 1996
(Show Context)
Citation Context ...ch networks uar-addressable. Distributed triangulation Predating the GNP algorithm, in the networking literature there were the IDMaps [9] of Francis et al., and several other beacon-based approaches =-=[17, 13, 20]-=- that used the triangle inequality to infer the distances that have not been measured. In particular, in [17, 13] the uv-distance duv is estimated by min(dub + dvb), where the minimum is taken over al... |

36 | On heterogeneous overlay construction and random node selection in unstructured p2p networks
- Vishnumurthy, Francis
- 2006
(Show Context)
Citation Context ... communicate with (and, in particular, measure distance to) any other node given its ID. Secondly, every node can select a node ID independently and uniformly at random among all nodes in the network =-=[21, 38]-=-. Such operation induces load on multiple nodes; to account for it, let us assume that when each node selects one random node ID, this induces a per-node load of (log n). We call such networks uar-add... |

30 | and B.Bhattacharjee. Finding close friends on the internet
- Kommareddy
- 2001
(Show Context)
Citation Context ...ch networks uar-addressable. Distributed triangulation Predating the GNP algorithm, in the networking literature there were the IDMaps [9] of Francis et al., and several other beacon-based approaches =-=[17, 13, 20]-=- that used the triangle inequality to infer the distances that have not been measured. In particular, in [17, 13] the uv-distance duv is estimated by min(dub + dvb), where the minimum is taken over al... |

30 | The black-box complexity of nearest-neighbor search
- Krauthgamer, Lee
- 2005
(Show Context)
Citation Context |

26 |
Property Testing (A Tutorial
- Ron
- 2001
(Show Context)
Citation Context ... also obtain similar guarantees for triangulations. We have already discussed the connection of our work to distance labeling. It is also related to the work on property testing in metric spaces (see =-=[35]-=- for a survey); however, this work considers problems quite different from what we study here, and makes use of different sampling models and objective functions. Finally, our use of triangulation cor... |

18 | A note on the nearest neighbor in growth-restricted metrics
- Hildrum, Kubiatowicz, et al.
- 2004
(Show Context)
Citation Context ...6, 23, 22, 19, 28, 4]. At the same time, several recent studies suggest the bounded growth rate of balls as a useful way to capture the structural properties of the Internet distance matrix (see e.g. =-=[34, 8, 30, 31, 41, 15]-=-). Our contributions: In this paper we settle a question left open by [19]: we show that there exists a fully distributed algorithm that embeds a doubling metric into a low-dimensional Euclidean space... |

14 |
Scale-free behavior of the internet global performance
- Percacci, Vespignani
- 2003
(Show Context)
Citation Context |

14 |
Bypassing the embedding: approximation schemes and compact representations for low dimensional metrics
- Talwar
- 2004
(Show Context)
Citation Context ...cally in [19]. We use our techniques to extend ɛ-relaxed embeddings and triangulations to infinite metrics and arbitrary measures, and to improve on the approximate distance labeling scheme of Talwar =-=[36]-=-. 1 Introduction A number of recent papers in the networking community study the distance matrix defined by the node-tonode latencies in the Internet 1 (e.g. [9, 13, 8, 20, 37]) and, in particular, pr... |

12 | Macroscopic Internet topology and performance measurements from the DNS root name servers
- Fomenkov, claffy, et al.
- 2001
(Show Context)
Citation Context |

6 | Course notes: Expander graphs and their applications. http://www.math.ias.edu/ boaz/ExpanderCourse - Linial, Wigderson - 2002 |

4 |
Predicting Internet network distances using peer-to-peer measurements
- PALM
- 2004
(Show Context)
Citation Context |

3 |
On hierarchical routing in bounded-growth metrics
- Chan, Gupta, et al.
- 2005
(Show Context)
Citation Context |

3 | The Evolution of Expander Graphs - Xiao - 2003 |