H-Pastry: An inter-domain topology aware overlay for the support of name-resolution services in the future Internet

Abstract

Abstract-Overlay networks are widely used for locating and disseminating information by means of custom routing and forwarding on top of an underlying network. Distributed Hash Table (DHT) based overlays in particular, provide good scalability and load balancing properties. However, these come at the cost of inefficient routing, caused by the lack of adaptation to the underlying network, as DHTs often overlook physical network proximity, administrative boundaries and/or inter-domain routing policies. In this paper we show how to construct a DHT-based overlay network that takes all these aspects into account, so as to ease the global deployment of Future Internet architectures which require large-scale name resolution, such as Information-Centric Networking (ICN) and the Internet of Things (IoT). Based on the Pastry distributed object location and routing substrate and the Canon paradigm for multi-level DHTs, we developed H-Pastry, an overlay DHT scheme that harvests the scalability and load balancing features of DHTs, while also adapting to the underlying network topology, administrative structure and routing policies. We evaluate the performance characteristics of the proposed scheme through an extensive set of detailed simulations over realistic inter-network topologies. Our results show that HPastry substantially improves routing by reducing both overlay path stretch (by up to 55%) and routing policy violations (by up to 70%), compared to the Canonical (multi-level) Chord DHT. In addition, the design of H-Pastry keeps traffic within administrative boundaries as far as possible, reducing interdomain hops by up to 27% compared to Pastry, while also creating excellent opportunities for the support of caching and multicast.

A Temporally Hierarchical Deployment Architecture for an Enhanced Name Resolution System

“…One popular approach to improve the scalability for flat name-based NRS is to use a DHT-based routing method to store the name bindings in appropriate resolvers via hash functions. Nested and hierarchical DHT architecture were utilized to implement distributed and scalable NRS based on flat names, such as Multi-level Distributed Hash Table (MDHT) [22], Hierarchical Distributed Hash Table (HDHT) [26], Scalable Multi-level Virtual Distributed Hash Table (SVMDHT) [27], and Hierarchical Pastry (H-Pastry) [28]. However, the administrative scope of the DHT mechanism is a prominent problem [20], and the existing methods have almost no specific solutions to solve the problem of hierarchical and nested area division and management.…”

“…In addition, our ENRS is also different from existing hierarchical and nested ICN NRS, e.g., DONA [9], HDHT [26], H-Pastry [28], and CURING [50], which organize the hierarchical and nested structure based on the hierarchical structure of the underlying inter-domain topology, i.e., at theAutonomous System (AS) level, the Point of Presence (POP) level, or the Access Node (AN) level. Our ENRS not only considers the underlying topology, but also quantifies the hierarchical and nested areas by transmission latency constraints from the user nodes to the serviced resolvers.…”

On the inter-domain scalability of route-by-name Information-Centric Network Architectures

“…However, the presented performance evaluation is based on an oversimplified abstraction of the inter-domain topology i.e., a full k-ary tree, overlooking the effect of multihoming and peering links. DHT-NRS [11] is based on H-Pastry [20], a multi-level version of the Pastry DHT [21] that tries to adapt to the underlying network hierarchy by taking physical network proximity, administrative domain boundaries and inter-domain routing policies into account. Although DHT-NRS exhibits better performance than other DHT-based NRSs, it reduces but does not eliminate the intrinsic problems of DHTs [11].…”

Towards Load Balance and Maintenance in a Structured P2P Network for Locality Sensitive Hashing

“…However, the uniform distribution of keys makes Chord unable to support range queries, which require data locality. Recently, several schemes have focused on similarity search over high-dimensional data in structured P2P overlay networks [6][7][8][9][10][11]. These schemes design different indexing structures in DHT-based P2P networks to support complicated similarity search.…”