## Meridian: A Lightweight Framework for Network Location without Virtual Coordinates (2005)

Venue: | In Proc. of ACM SIGCOMM |

Citations: | 1 - 0 self |

### BibTeX

@INPROCEEDINGS{Wong05meridian:a,

author = {Bernard Wong and Aleksandrs Slivkins and Emin Gün Sirer},

title = {Meridian: A Lightweight Framework for Network Location without Virtual Coordinates},

booktitle = {In Proc. of ACM SIGCOMM},

year = {2005}

}

### OpenURL

### Abstract

Selecting nodes based on their position in the network is a basic building block for many distributed systems. This paper describes a peer-to-peer overlay network for performing position-based node selection. Our system, Meridian, provides a lightweight, accurate and scalable framework for keeping track of location information for participating nodes. The framework consists of an overlay network structured around multi-resolution rings, query routing with direct measurements, and gossip protocols for dissemination. We show how this framework can be used to address three commonly encountered problems in large-scale distributed systems without having to compute absolute coordinates; namely, closest node discovery, central leader election, and locating nodes that satisfy target latency constraints. We show analytically that the framework is scalable with logarithmic convergence when Internet latencies are modeled as a growth-constrained metric, a low-dimensional Euclidian metric, or a metric of low doubling dimension. Large scale simulations, based on latency measurements from 6.25 million node-pairs, and an implementation deployed on PlanetLab both show that the framework is accurate and effective. 1

### Citations

3473 | Chord: A scalable peer-to-peer lookup service for internet applications
- Stoica, Morris, et al.
- 2001
(Show Context)
Citation Context ...cates the closest node by performing a multi-hop search where each hop exponentially reduces the distance to the target. This is similar to searching in structured peer-to-peer networks such as Chord =-=[50]-=-, Pastry [45] and Tapestry [54], where each hop brings the query exponentially closer to the destination, though in Meridian the routing is performed using physical latencies instead of numerical dist... |

2700 | A scalable content-addressable network
- Ratnasamy, Francis, et al.
- 2001
(Show Context)
Citation Context ..., an essential requirement in large-scale networks, requires additional mechanisms besides virtual coordinates. Peer-to-peer substrates that can naturally work with Euclidian coordinates, such as CAN =-=[43]-=- and P-Trees [10], can reduce the state requirements per node; however, both systems introduce substantial complexity and bandwidth overhead in addition to the overhead of network embedding. And our s... |

1876 | P.: Randomized Algorithms - Motwani, Raghavan - 1995 |

1697 | Pastry: Scalable, distributed object location and routing for large-scale peer-to-peer systems
- Rowstron, Druschel
- 2001
(Show Context)
Citation Context ...sest node by performing a multi-hop search where each hop exponentially reduces the distance to the target. This is similar to searching in structured peer-to-peer networks such as Chord [50], Pastry =-=[45]-=- and Tapestry [54], where each hop brings the query exponentially closer to the destination, though in Meridian the routing is performed using physical latencies instead of numerical distances in asFi... |

1102 | A Case for End System Multicast - Chu, Rao, et al. - 2000 |

1039 | Tapestry: An infrastructure for fault-tolerant wide-area location and routing
- Zhao, Kubiatowicz, et al.
- 2001
(Show Context)
Citation Context ...rming a multi-hop search where each hop exponentially reduces the distance to the target. This is similar to searching in structured peer-to-peer networks such as Chord [50], Pastry [45] and Tapestry =-=[54]-=-, where each hop brings the query exponentially closer to the destination, though in Meridian the routing is performed using physical latencies instead of numerical distances in asFigure 3: Central le... |

991 | Resilient overlay networks - Andersen - 2001 |

847 | Widearea cooperative storage with CFS - Dabek, Kaashoek, et al. - 2001 |

701 |
Epidemic algorithms for replicated database maintenance
- Demers, Greene, et al.
- 1987
(Show Context)
Citation Context ...ode discovery allows the Meridian overlay to be loosely connected, highly robust and inexpensively kept up-to-date of membership changes. Our gossip protocol is based on an anti-entropy push protocol =-=[14]-=- that implements a membership service. The central goal of our gossip protocol is not for each node to discover every node in the system, but simply for each node to discover a sufficiently diverse se... |

665 | Storage management and caching in PAST, a large-scale, persistent peer-to-peer storage utility - Rowstron, Druschel - 2001 |

528 | Predicting Internet network distance with coordinates-based approaches
- Ng, Zhang
(Show Context)
Citation Context ... network metric is to perform a network embedding, that is, to map high-dimensional network measurements into a location in a smaller Euclidian space. For instance, recent work in network positioning =-=[40, 12, 36, 52, 48, 41, 9, 39, 35]-=- maps a large vector of node-to-node latency measurements on the Internet into a single point in a ¢ - dimensional space. The resulting embedded address facilitates location-aware node selection. Whil... |

474 | Accessing nearby copies of replicated objects in a distributed environment
- Plaxton, Rajaraman, et al.
- 1997
(Show Context)
Citation Context ...ics and doubling metrics. In particular, growth-constrained metrics have been used as a reasonable abstraction of Internet latencies in the analysis of the object-location algorithm of Plaxton et al. =-=[42]-=-. Using a more general family of doubling metrics leads to good guarantees even for metrics that combine very dense and very sparse regions. We focus on the case when the rate of churn and fluctuation... |

469 | Vivaldi: A decentralized network coordinate system
- Dabek, Cox, et al.
- 2004
(Show Context)
Citation Context ... network metric is to perform a network embedding, that is, to map high-dimensional network measurements into a location in a smaller Euclidian space. For instance, recent work in network positioning =-=[40, 12, 36, 52, 48, 41, 9, 39, 35]-=- maps a large vector of node-to-node latency measurements on the Internet into a single point in a ¢ - dimensional space. The resulting embedded address facilitates location-aware node selection. Whil... |

294 | Topologically-aware overlay construction and server selection
- Ratnasamy, Handley, et al.
- 2002
(Show Context)
Citation Context ...he majority of routes, and the choice of an appropriate � is not obvious without prior knowledge of the node distribution. Another landmark based technique for closest node discovery is described in =-=[44]-=-, where each node determines its bin number via measurements to well known landmarks. A node wishing to find its closest node determines its own bin number, queries a modified DNS server for other nod... |

266 | The end-to-end effects of internet path selection - SAVAGE, COLLINS, et al. - 1999 |

256 | IDMaps: A Global Internet Host Distance Estimation Service
- Francis, Jamin, et al.
(Show Context)
Citation Context ...he potential energy of the spring also minimizes the embedding error. BBS [48] performs a similar simulation to calculate coordinates, simulating an explosion of particles under a force field. IDMaps =-=[16]-=-, like network embedding systems, can compute the approximate distance between two IP addresses without direct measurement based on strategically placed tracer nodes. IDMaps incurs inherent errors bas... |

214 | Operating system support for planetary-scale network services
- Bavier, Bowman, et al.
- 2004
(Show Context)
Citation Context ...rformance of our system is not an artifact of our measurements. We evaluate Meridian through simulation parameterized by a large-scale network measurement study, and through a deployment on PlanetLab =-=[2]-=-. For our measurement study, we collected node-to-node round-trip latency measurements for 2500 nodes and 6.25 million node pairs on the Internet using the King [17] measurement technique. We use 500 ... |

168 | Distributed object location in a dynamic network
- Hildrum, Kubiatowicz, et al.
- 2002
(Show Context)
Citation Context ...at of discovering the closest node to a targeted reference point. This is a basic operation in content distribution networks (CDNs) [25], large-scale multiplayer games [34], and peer-to-peer overlays =-=[24, 26, 6, 5]-=-. Having the closest node serve the client or operate on the target can significantly reduce response time and aggregate network load. For instance, a geographically distributed peer-topeer web crawle... |

163 |
Estimating Latency between Arbitrary Internet End Hosts
- King
- 2002
(Show Context)
Citation Context ...d through a deployment on PlanetLab [2]. For our measurement study, we collected node-to-node round-trip latency measurements for 2500 nodes and 6.25 million node pairs on the Internet using the King =-=[17]-=- measurement technique. We use 500 of these nodes as targets, and the remaining 2000 as the overlay nodes in our experiments. Overall, this paper makes three contributions. First, it outlines a lightw... |

162 | PIC: Practical Internet coordinates for distance estimation
- Costa, Castro, et al.
- 2004
(Show Context)
Citation Context ... network metric is to perform a network embedding, that is, to map high-dimensional network measurements into a location in a smaller Euclidian space. For instance, recent work in network positioning =-=[40, 12, 36, 52, 48, 41, 9, 39, 35]-=- maps a large vector of node-to-node latency measurements on the Internet into a single point in a ¢ - dimensional space. The resulting embedded address facilitates location-aware node selection. Whil... |

162 | Virtual landmarks for the Internet
- Tang, Crovella
- 2003
(Show Context)
Citation Context |

153 | J.R.: Bounded geometries, fractals, and lowdistortion embeddings
- Gupta, Krauthgamer, et al.
- 2003
(Show Context)
Citation Context ...allsuch that ball� � ¤���¨ every can be by� covered balls of ����� radius . Metrics of low doubling dimension is a strictly more general family than growth-constrained metrics since it is easy to see =-=[18]-=- that DIM is at most four times the KR-dimension, but the converse is not true, e.g. for a ��� ��� �©��������� subset of the real line ��� DIM , but KR-dimension � � is . Intuitively, doubling metrics... |

149 | Locating nearby copies of replicated internet servers
- Guyton, Schwarz
- 1995
(Show Context)
Citation Context ...no other nodes belong in the same bin, and chooses a random node from the retrieved set of servers. Several different proactive techniques to locate the closest replica to the client are evaluated in =-=[19]-=-. These techniques offer different methods to construct a connectivity graph by means of polling the routing table of connecting hops, explicitly sending routing probes, or limited probing with triang... |

149 | M.: Finding nearest neighbors in growth-restricted metrics
- Karger, Ruhl
- 2002
(Show Context)
Citation Context ...at of discovering the closest node to a targeted reference point. This is a basic operation in content distribution networks (CDNs) [25], large-scale multiplayer games [34], and peer-to-peer overlays =-=[24, 26, 6, 5]-=-. Having the closest node serve the client or operate on the target can significantly reduce response time and aggregate network load. For instance, a geographically distributed peer-topeer web crawle... |

145 | Lighthouses for scalable distributed location
- Pias, Crowcroft, et al.
- 2003
(Show Context)
Citation Context |

141 | Spatial gossip and resource location protocols
- Kempe, Kleinberg, et al.
(Show Context)
Citation Context ... of Internet latencies in past work (see the intro of [22] for a short survey). Growth-constrained metrics have also been considered in the context of dimensionality in graphs [32] and spatial gossip =-=[27]-=-. We start with a model where the metric on the Meridian nodes is growth-constrained, but we make no such assumption about the non-Meridian nodes. This is important because even in an unfriendly metri... |

133 | A novel server selection technique for improving the response time of a replicated service
- Fei, Bhattacharjee, et al.
- 1998
(Show Context)
Citation Context ...ent based on strategically placed tracer nodes. IDMaps incurs inherent errors based on the client’s distance to its closest tracer server and requires deploying system wide infrastructure. Other work =-=[15]-=- has also examined how to delegate probing to specialized nodes in the network. There has also been theoretical work [28, 49] on explaining the empirical success of network embeddings and IDMaps-style... |

133 | Big-bang simulation for embedding network distances in Euclidean space
- Shavitt, Tankel
(Show Context)
Citation Context |

129 | Exploiting network proximity in peer-to-peer overlay networks
- Castro, Druschel, et al.
- 2002
(Show Context)
Citation Context ...at of discovering the closest node to a targeted reference point. This is a basic operation in content distribution networks (CDNs) [25], large-scale multiplayer games [34], and peer-to-peer overlays =-=[24, 26, 6, 5]-=-. Having the closest node serve the client or operate on the target can significantly reduce response time and aggregate network load. For instance, a geographically distributed peer-topeer web crawle... |

128 | Server selection using dynamic path characterization in wide-area networks
- Carter, Crovella
- 1997
(Show Context)
Citation Context ...or limited probing with triangulation. The study assumes the network conditions and topology remain relatively static, and does not directly address scalability. Dynamic server selection was found in =-=[3]-=- to be more effective than static server selection due to the variability of route latency over time and the large divergence between hop count and latency. Simulations [4] using a simple dynamic serv... |

123 | Navigating nets: Simple algorithms for proximity search
- Krauthgamer, Lee
(Show Context)
Citation Context ... � a -dimensional Euclidean metric ��£¥¤���¨ DIM . Doubling dimension has been introduced in the mathematical literature (see [20]) and has recently become a hot topic in the theoretical CS community =-=[18, 30, 33, 51, 28, 49]-=-; in particular it was used to model Internet latencies in the context of distributed algorithms for embedding and distance estimation [28, 49]. For metrics of low doubling dimension, well-formed ring... |

120 |
Lectures on analysis on metric spaces
- Heinonen
- 2001
(Show Context)
Citation Context ...ional Euclidean metrics; it is known [18] that for any finite point set in � a -dimensional Euclidean metric ��£¥¤���¨ DIM . Doubling dimension has been introduced in the mathematical literature (see =-=[20]-=-) and has recently become a hot topic in the theoretical CS community [18, 30, 33, 51, 28, 49]; in particular it was used to model Internet latencies in the context of distributed algorithms for embed... |

114 | The measured performance of content distribution networks
- Johnson, Carr, et al.
- 2001
(Show Context)
Citation Context ... virtual coordinates. The first, and most significant, problem is that of discovering the closest node to a targeted reference point. This is a basic operation in content distribution networks (CDNs) =-=[25]-=-, large-scale multiplayer games [34], and peer-to-peer overlays [24, 26, 6, 5]. Having the closest node serve the client or operate on the target can significantly reduce response time and aggregate n... |

109 | A network positioning system for the internet
- Ng, Zhang
- 2004
(Show Context)
Citation Context |

83 | Measured descent: A new embedding method for finite metrics
- Krauthgamer, Lee, et al.
(Show Context)
Citation Context ... � a -dimensional Euclidean metric ��£¥¤���¨ DIM . Doubling dimension has been introduced in the mathematical literature (see [20]) and has recently become a hot topic in the theoretical CS community =-=[18, 30, 33, 51, 28, 49]-=-; in particular it was used to model Internet latencies in the context of distributed algorithms for embedding and distance estimation [28, 49]. For metrics of low doubling dimension, well-formed ring... |

78 |
Triangulation and embedding using small sets of beacons
- Kleinberg, Slivkins, et al.
(Show Context)
Citation Context ... � a -dimensional Euclidean metric ��£¥¤���¨ DIM . Doubling dimension has been introduced in the mathematical literature (see [20]) and has recently become a hot topic in the theoretical CS community =-=[18, 30, 33, 51, 28, 49]-=-; in particular it was used to model Internet latencies in the context of distributed algorithms for embedding and distance estimation [28, 49]. For metrics of low doubling dimension, well-formed ring... |

53 | Preserving peer replicas by rate-limited sampled voting - Maniatis, Rosenthal, et al. - 2003 |

50 | Querying peer-to-peer networks using p-trees
- Crainiceanu, Linga, et al.
- 2004
(Show Context)
Citation Context ...quirement in large-scale networks, requires additional mechanisms besides virtual coordinates. Peer-to-peer substrates that can naturally work with Euclidian coordinates, such as CAN [43] and P-Trees =-=[10]-=-, can reduce the state requirements per node; however, both systems introduce substantial complexity and bandwidth overhead in addition to the overhead of network embedding. And our simulation results... |

45 | Proximity neighbor selection in tree-based structurd peer-to-peer overlays
- Castro, Druschel, et al.
- 2003
(Show Context)
Citation Context |

33 | Backup path allocation based on a correlated link failure probability model in overlay networks - CUI, STOICA, et al. |

30 | Distributed Approaches to Triangulation and Embedding
- Slivkins
- 2004
(Show Context)
Citation Context |

28 | The black-box complexity of nearest-neighbor search - Krauthgamer, Lee |

28 | Introspective failure analysis: Avoiding correlated failures in peer-to-peer systems - Weatherspoon, Moscovitz, et al. - 2002 |

25 |
Pcoord: Network position estimation using peer-to-peer measurements
- Lehman, Lerman
- 2004
(Show Context)
Citation Context |

21 | The intrinsic dimensionality of graphs
- Krauthgamer, Lee
- 2003
(Show Context)
Citation Context ...a reasonable abstraction of Internet latencies in past work (see the intro of [22] for a short survey). Growth-constrained metrics have also been considered in the context of dimensionality in graphs =-=[32]-=- and spatial gossip [27]. We start with a model where the metric on the Meridian nodes is growth-constrained, but we make no such assumption about the non-Meridian nodes. This is important because eve... |

17 | On the Network Impact of Dynamic Server Selection.ComputerNetworks
- Carter, Crovella
- 1999
(Show Context)
Citation Context ...rver selection was found in [3] to be more effective than static server selection due to the variability of route latency over time and the large divergence between hop count and latency. Simulations =-=[4]-=- using a simple dynamic server selection policy, where all replica servers are probed and the server with the lowest average latency is selected, show the positive system wide effects of latency-based... |

13 |
Bypassing the Embedding: Approximation Schemes and Compact Representations for Growth Restricted Metrics
- Talwar
(Show Context)
Citation Context |

12 |
Object location in realistic networks
- Hildrum, Krauthgamer, et al.
- 2004
(Show Context)
Citation Context ... and good load balance. This analysis is general and applies to Internet latencies that cannot be accurately modeled with a Euclidean metric. Following a line of previous work on object location (see =-=[21]-=- for a recent summary), we give guarantees for the family of growthconstrained metrics. Moreover, we support a much wider family of metrics of low doubling dimension which has recently become popular ... |

10 |
A note on finding the nearest neighbor in growth-restricted metrics
- Hildrum, Kubiatowicz, et al.
- 2004
(Show Context)
Citation Context ...al grid the KR-dimension is £¥¤���¨ , growth-constrained metrics can be seen as generalized grids; they have been used as a reasonable abstraction of Internet latencies in past work (see the intro of =-=[22]-=- for a short survey). Growth-constrained metrics have also been considered in the context of dimensionality in graphs [32] and spatial gossip [27]. We start with a model where the metric on the Meridi... |

8 | Another way to find the nearest neighbor in growth-restricted metrics - Hildrum, Kubiatowicz, et al. - 2003 |