Towards new load-balancing schemes for structured peer-to-peer grids

Abstract

In this paper, we introduce the concept of a structured peer-to-peer grid and present our contribution to this new world by means of Dermi, a wide-area decentralized object middleware. In addition, we focus on the design and implementation of a load-balancing facility by using the functionalities provided by this middleware. We present two different approaches to achieve load-balancing in our system: a completely decentralized solution by using the anycall abstraction, and a distributed interceptor-based one. Each of them can be used in a wide variety of scenarios, depending on our needs.

Introduction

Grid and Peer-to-Peer (P2P) Networks are two concepts, which may seem different at first sight. There exist no exact definitions that allow us to compare their similarities and differences in an unambiguous way. Nevertheless, the general idea seems to split both worlds into those cases where infrastructure is used to allow seamless access to supercomputers and their datasets (Grids), and those which enable ad hoc communities of low-end clients to advertise and access the files on the communal computers (P2P).

Trying to merge the best of both worlds, the concept of a P2P Grid was devised [1]. A P2P Grid contains a set of services including the services of Grids and P2P networks and support naturally environments having features of both limiting cases. In P2P Grids’ architecture, Web services play a very important role. In addition, there is an event service, which links these Web services and other resources together. In P2P Grids, everything is a resource, and they are exposed directly to users and to other services. An important feature is that these entities are built as distributed objects, which are constructed as services allowing their properties and methods to be accessed by a message-based protocol.

Normally, these P2P Grids have been built on top of unstructured P2P networks. This kind of networks at the extreme case are exemplified by the Gnutella-like systems, where all participants join the network in an anarchic way, and messages are flooded all over the network (normally messages are only sent a few hops away from the origin, thus being extremely inefficient when searching for a precise resource). Other alternative techniques to flooding include the use of Random Walkers, or expanding ring techniques. However, they do not elegantly solve the problem of resource location in a deterministic time slot, being it impossible to know how long it will take to find a concrete resource, if it is found.

Many research efforts have lately been directed towards building structured P2P networks, also called overlay routing networks, which are the base for constructing distributed hash tables (DHTs). These systems provide a self-organizing substrate ideally suited for large-scale P2P applications. A wide variety of decentralized services can be built on top of these overlays. Examples include application-level multicast, network storage, and content distribution. Being structured overlays, applications can locate any resource in a small number of network hops (typically O(log n)), and they require small-sized per-node routing tables.

Relating P2P Grids with structured peer-to-peer overlay networks, we have developed a structured P2P Grid middleware platform called Decentralized Event Remote Method Invocation (Dermi) [6]. Dermi is a distributed event-based object middleware built on top of a DHT-based P2P overlay network. It is a first step towards building a complete P2P Grid middleware platform, providing many services. Dermi provides a solid building block for the development of wide-area distributed applications. It implements many of the functionalities required for a P2P Grid, including an event service which glues the distributed objects representing resources.

In this paper we introduce the term of structured P2P Grid, which we define as a P2P Grid whose network substrate is based on a structured DHT approach. This means that the P2P Grid itself is built on top of a Key-Based Routing (KBR) infrastructure. The main advantage of this approach is that these kinds of Grids can benefit from better message routing and its inherent locality properties.

The main contribution of this paper is describing two load-balancing mechanisms, which we have designed by using the different services provided by Dermi. We demonstrate the viability of both schemes by performing simulations and experimental evaluations.

The paper is structured as follows: we start in Section 2 with a brief description of Dermi and the main services it provides. Section 3 describes the two mechanisms we have designed to achieve load-balancing using Dermi. The following section contains simulations and experimental evaluations which validate both approaches. Next, we discuss the contributions of Dermi to the P2P Grid world and continue exploring existent work related to our system. Finally, we present a summary and an outline of future research.