K. Hildrum, J. Kubiatowicz, S. Rao, and B. Y. Zhao. Distributed data location in a dynamic network. In Proceedings of the 14th Annual ACM Symposium on Parallel Algorithms and Architectures, pages 41--52, August 2002.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the object and the name of the object. In Riptide, we use SHA 1 [25] but the system has been designed to permit the use of any hashing algorithm. 3. 2 Distributed Object Location and Routing Peer to peer researchers have begun to explore dis tributed object location and routing (DOLR) services [17, 28, 35, 37]. DOLR systems are overlay net works that offer a distributed framework in which objects that are named by GUIDs can be located quickly. Since information about these objects is distributed throughout the system, messages are routed to objects by passing them from node to node until they reach ....

....in which objects that are named by GUIDs can be located quickly. Since information about these objects is distributed throughout the system, messages are routed to objects by passing them from node to node until they reach their destination. OceanStore is built on top of the Tapestry DOLR [17, 50]. OceanStore also uses attenuated Bloom filters [29] for probabilistic local area routing. The combination of Tapestry and attenuated Bloom filters provides locality in routing. If a DOLR is said to exhibit good locality, then messages will be routed over a minimal overlay path to their ....

[Article contains additional citation context not shown here]

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Zhao. Distributed data location in a dynamic network. In Proc. of ACM SPAA, 2002.

....the object and the name of the object. In Riptide, we use SHA 1 [25] but the system has been designed to permit the use of any hashing algorithm. 3. 2 Distributed Object Location and Routing Peer to peer researchers have begun to explore distributed object location and routing (DOLR) services [17, 28, 35, 37]. DOLR systems are overlay networks that o er a distributed framework in which objects that are named by GUIDs can be located quickly. Since information about these objects is distributed throughout the system, messages are routed to objects by passing them from node to node until they reach their ....

....in which objects that are named by GUIDs can be located quickly. Since information about these objects is distributed throughout the system, messages are routed to objects by passing them from node to node until they reach their destination. OceanStore is built on top of the Tapestry DOLR [17, 50]. OceanStore also uses attenuated Bloom lters [29] for probabilistic local area routing. The combination of Tapestry and attenuated Bloom lters provides locality in routing. If a DOLR is said to exhibit good locality, then messages will be routed over a minimal overlay path to their destination, ....

[Article contains additional citation context not shown here]

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Zhao. Distributed data location in a dynamic network. In Proc. of ACM SPAA, 2002.

....for join and leave operations. Maintaining the PRR neighbor pointers, however, is a nontrivial task, especially if the distance function is changing, or if nodes are frequently joining or leaving the network. Although recent research results have reduced the restriction on the metric spaces [11, 12], providing similar locality properties on general metric spaces remains an open problem. When applied to arbitrary metric spaces, some performance bounds established by PRR are considerably weakened. For example, on growth restricted metric spaces, the in degree of a node is O(log n) expected ....

....needs to set d = lg n, which may not be feasible without a good anticipation of n or if n changes dramatically during the lifetime of the network. The importance of locality is now widely recognized and most name resolution schemes go to significant lengths to exploit locality, be it rigorously [11, 12, 20] or heuristically [4, 24, 28] As discussed in Section 1, there is a tradeoff between simplicity and effectiveness of exploiting locality, and SPRR attempts to exploit locality without sacrificing simplicity. At a high level, SPRR, especially with the implementation maintaining only predecessors ....

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Y. Zhao. Distributed data location in a dynamic network. In Proceedings of the 14th Annual ACM Symposium on Parallel Algorithms and Architectures, pages 41--52, August 2002. 19

....Internet Data Centers (IDC) of major ISPs with good connectivity to the backbone. Each SCAN server may contain replicas for a variety of data items. One novel aspect of the SCAN system is that it assumes SCAN servers participate in a distributed routing and location (DOLR) system, called Tapestry [11]. Tapestry permits clients to locate nearby replicas without global communication. There are three types of data illustrated in Figure 1: Data sources and replicas are the primary topic of this paper and reside on SCAN servers. Caches are the images of data that reside on clients and are beyond ....

.... connections [8, 4, 6, 13, 28] Most ALM systems utilize a central node to maintain state for all existing children [6, 13, 19, 4] and are not very scalable; some replicate the root to help with this problem [13] Similar to SCAN, Bayeux [28] is an overlay multicast system build on the Tapestry [11] DOLR; however, unlike SCAN, Bayeux filters all join requests through a replicated set of root nodes. Scribe [24] provides a scalable, eventnotification multicast system built on top of the Pastry [23] DOLR. It provides mechanisms for dynamic, decentralized construction of multicast groups, but ....

[Article contains additional citation context not shown here]

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Zhao. Distributed data location in a dynamic network. In Proc. of ACM SPAA, 2002.

....for structured peer to peer overlays which we use as a framework for measurement. Such a model attempts to capture the characteristics of an overlay which are of interest to an application writer. In Section 3, we describe our benchmarking environment, consisting of the Chord [13] and Tapestry [5, 16] implementations running on the PlanetLab testbed. In Section 4 we discuss two benchmarks for overlays, and the results of running them against our Chord and Tapestry deployments. 2 A Common Service Model Before describing our benchmarking work, we present a generic service model for structured ....

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Zhao. Distributed data location in a dynamic network. In Proc. of ACM SPAA, 2002.

....for structured peerto peer overlays which we use as a framework for measurement. Such a model attempts to capture the characteristics of an overlay which are of interest to an application writer. In Section 3, we describe our benchmarking environment, consisting of the Chord [13] and Tapestry [5, 16] implementations running on the PlanetLab testbed. In Section 4 we discuss two benchmarks for overlays, and the results of running them against our Chord and Tapestry deployments. In Section 5 we draw some conclusions and discuss future work. 2 A Common Service Model Before describing our ....

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Zhao. Distributed data location in a dynamic network. In Proc. of ACM SPAA, 2002.

....we discuss the Set Creator. To create dissemination sets, the Set Creator must first collect information about a sufficiently large set of nodes. The Creator relies on the properties of a Decentralized Object Location and Routing (DOLR) system such as CAN [15] Chord [18] Pastry [7] or Tapestry [8], to discover new nodes. A DOLR routes messages to nodes that most closely matches the requested destination. Thus, even though the DOLR address space is sparse, the Set Creator can use a scan of the address space to reach a random set of nodes. Servers that are willing to host a specified ....

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Zhao. Distributed data location in a dynamic network. In Proc. of ACM SPAA, 2002.

....heartbeats and continuous system repair. 1. Introduction Recent peer to peer systems have adopted decentralized object location and routing (DOLR) infrastructures to assist in organizing and manipulating their data. Prominent examples of DOLRs include CAN[6] Chord[8] Pastry[2] Tapestry[3, 11], and other Plaxton, Rajaraman, Richa[5] structures. DOLRs provide sufficient probabilistic routing guarantees to find an object if it exists; but not enough, if any, reliability guarantees. This dilemma is often solved with replication[1, 2] or other forms of redundancy[4, 9] Unfortunately, ....

....Redundant Links The problem with object heartbeats is that they traverse redundant links. That is, the number of stored objects is much greater than the number of neighbor links. Most of the heartbeats can be aggregated together. 3. Location Independent Routing We now describe Tapestry[3, 11], a routing and location system. Tapestry is an IP overlay network that uses a distributed, fault tolerant architecture to track the location of every object in the network. Tapestry has two components: a routing mesh and a distributed directory service. Routing Mesh: Figure 1 shows a portion of ....

[Article contains additional citation context not shown here]

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Zhao. Distributed data location in a dynamic network. In Proc. of ACM SPAA, 2002.

....Internet Data Centers (IDC) of major ISPs with good connectivity to the backbone. Each SCAN server may contain replicas for a variety of data items. One novel aspect of the SCAN system is that it assumes SCAN servers participate in a distributed routing and location (DOLR) system, called Tapestry [11]. Tapestry permits clients to locate nearby replicas without global communication. There are three types of data illustrated in Figure 1: Data sources and replicas are the primary topic of this paper and reside on SCAN servers. Caches are the images of data that reside on clients and are beyond ....

.... connections [8, 4, 6, 13, 28] Most ALM systems utilize a central node to maintain state for all existing children [6, 13, 19, 4] and are not very scalable; some replicate the root to help with this problem [13] Similar to SCAN, Bayeux [28] is an overlay multicast system build on the Tapestry [11] DOLR; however, unlike SCAN, Bayeux lters all join requests through a replicated set of root nodes. Scribe [24] provides a scalable, eventnoti cation multicast system built on top of the Pastry [23] DOLR. It provides mechanisms for dynamic, decentralized construction of multicast groups, but ....

[Article contains additional citation context not shown here]

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Zhao. Distributed data location in a dynamic network. In Proc. of ACM SPAA, 2002.

No context found.

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Y. Zhao. Distributed data location in a dynamic network. In Proceedings of the 14th Annual ACM Symposium on Parallel Algorithms and Architectures, pages 41--52, August 2002.

No context found.

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Y. Zhao. Distributed data location in a dynamic network. In Proceedings of the 14th Annual ACM Symposium on Parallel Algorithms and Architectures, pages 41--52, August 2002.

No context found.

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Y. Zhao. Distributed data location in a dynamic network. In Proceedings of the 14th Annual ACM Symposium on Parallel Algorithms and Architectures, pages 41--52, August 2002.

No context found.

Hildrum, K., Kubiatowicz, J., Rao, S., and Zhao, B. Distributed data location in a dynamic network. In Proc. of ACM SPAA (2002).

No context found.

K. Hildrum, J. Kubiatowicz, S. Rao, and B. Y. Zhao. Distributed data location in a dynamic network. In Proceedings of the 14th Annual ACM Symposium on Parallel Algorithms and Architectures, pages 41--52, August 2002.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC