Simulating and Optimizing a Peer-to-Peer Computing Framework

Parallel Computing, 2007

This paper presents a Peer-to-Peer (P2P) infrastructure that supports a large scale grid. The P2P infrastructure is implemented in Java and federates Java Virtual Machines (JVMs) for computation. The management of shared JVMs is decentralized, self-organized, and configurable.

In this technical report, we present our findings on network simulators, which can be used to help test and design a P2P system. The Portal-based P2P system under development uses portals to provide a user interface and P2P volunteers to provide resources to the network. The aim is to create an infrastructure that can be used by the scientific community, based on existing social networks, to help download and maintain large datasets. Network simulators provide a virtual network environment, which can potentially provide assurance for software developers, when making design decisions. This report outlines the architecture of the Portal-based P2P system and defines a set of criteria for a suitable simulator, to aid the development of the P2P system. A number of network simulators are reviewed and briefly tested for their suitability based on criteria. The experience gained from reviewing network simulators is presented in this report. 2

Proceedings of The …, 2006

As scientists, our research must be tested and evaluated in order for it to be shown to be valid. Part of this process includes providing reproducible results so that peers are able to confirm any findings. The techniques used to achieve this include analytical solutions, simulators and experiments with the actual system, however in the area of Peer-to-Peer (P2P) computing, this gives rise to a number of challenges.

2012

Peer-to-Peer (P2P) computing (also called 'public-resource computing') is an effective approach to perform computation of large tasks. Currently used P2P computing systems (e.g., BOINC) are most often centrally managed, i.e., the final result of computations is created at a central node using partial results -what may be not efficient in the case when numerous participants are willing to download the final result. In this paper, we propose a novel approach to P2P computing systems. We assume that results can be delivered to all peers in a distributed way using three types of network flows: unicast, Peer-to-Peer and anycast. We describe our concept of the system architecture with a special focus on the decision strategies developed for system participants. Using our discrete realtime simulator we evaluate the proposed strategies in various scenarios and present a comprehensive analysis of obtained results. According to obtained results, the Peer-to-Peer flow provides lower operational cost of the computing system compared to unicast and anycast flows. Moreover, an appropriate selection of decision strategy can significantly reduce the operational cost.

Concurrency and Computation: Practice and Experience, 2011

The inherent complex nature of current distributed computing architectures hinders the widespread adoption of these systems for mainstream use. In general, users have access to a highly heterogeneous set of compute resources, which may include clusters, grids, desktop grids, clouds, and other compute platforms. This heterogeneity is especially problematic when running parallel and distributed applications. Software is needed which easily combines as many resources as possible into one coherent computing platform.

The interesting properties of P2P systems (high availability despite node volatility, support for heterogeneous architectures, high scalability, etc.) make them attractive for distributed computing. However, conducting large-scale experiments with these systems arise as a major challenge. Simulation allows to model only partially the behavior of P2P prototypes. Experiments on real testbeds encounter serious difficulty with large-scale deployment and control of peers. This paper shows that using an optimized version of the JXTA Distributed Framework (JDF) allows to easily deploy, configure and control P2P experiments. We illustrate these features with sample tests performed with our JUXMEM JXTA-based grid data sharing service, for various large-scale configurations. Unité de recherche INRIA Rennes IRISA, Campus universitaire de Beaulieu, 35042 RENNES Cedex (France) Téléphone : 02 99 84 71 00 -International : +33 2 99 84 71 00 Télécopie : 02 99 84 71 71 -International : +33 2 99 84 71 71 Vers la grande échelle dans les expériences P2P: une approche basée sur JXTA Distributed Framework Résumé : Les propriétés intéressantes des systèmes pair-à-pair (haute disponibilité malgré la volatilité des noeuds, support des architectures hétérogènes, passage à l'échelle, etc.) les rendent séduisants pour les systèmes distribués. Toutefois, mener à bien des expériences à grande échelle pour ces systèmes apparaît comme un défi majeur. La simulation ne permet de modéliser que partiellement le comportement des prototypes pair-à-pair. Les expériences menées sur des plates-formes physiques font, elles, face à de sérieuses difficultés pour le déploiement et le contrôle à grande échelle de pairs. Ce papier montre que l'utilisation d'une version optimisée de JXTA Distributed Framework (JDF) permet de facilement déployer, configurer et contrôler des expériences pair-à-pair. Nous illustrons ces fonctionnalitées par des tests réalisés à différentes grandes échelles sur JUXMEM, notre service de partage de données pour la grille basé sur JXTA. Mots-clé : Pair-à-pair, grande échelle, expériences, JXTA Going Large-scale in P2P Experiments Using the JXTA Distributed Framework 3

akamaiuniversity.us

Currently, both Peer-to-Peer Computing (P2P) and Grid Computing have remained the most vibrant and useful forms of distributed computing all over the world. Their applications are such that they cut across both academia and industry. It has come to the notice of researchers that there are great misunderstanding and misinterpretation on what these forms of distributed computing actually portend and stand for. In this paper therefore, we take a critical look at comparative study of both computing technologies with aim of making readers understand in a clear cut what each really stands for. To have a good comparison, we start by giving a well referenced definition of Grid Computing as well as Peer-to-Peer Computing. Also, we used technical issues and general features in our comparison vis-à-vis the architecture, security issue, data movement, application deployment, and operating system requirement. We also considered the strength of both distributed computing system and finally we considered what could be the future of both technologies.

2015

We present an architecture for a fully decentralized peer-to-peer collaborative computing platform, offering services similar to Cloud Service Provider's Platform-as-a-Service (PaaS) model, using volunteered resources rather than dedicated resources. This thesis is motivated by three research questions: (1) Is it possible to build a peer-to-peer collaborative system using a fully decentralized infrastructure relying only on volunteered resources?, (2) How can light virtualization be used to mitigate the complexity inherent to the volunteered resources?, and (3) What are the minimal requirements for a computing platform similar to the PaaS cloud computing platform? We propose an architecture composed of three layers: the Network layer, the Virtual layer, and the Application layer. We also propose to use light virtualization technologies, or containers, to provide a uniform abstraction of the contributing resources and to isolate the host environment from the contributed environment. Then, we propose a minimal API specification for this computing platform, which is also applicable to PaaS computing platforms. The findings of this thesis corroborate the hypothesis that peer-to-peer collaborative systems can be used as a basis for developing volunteer cloud computing infrastructures. We outline the implications of using light virtualization as an integral virtualization primitive in public distributed computing platform. Finally, this thesis lays out a starting point for most volunteer cloud computing infrastructure development effort, because it circumscribes the essential requirements and presents solutions to mitigate the complexities inherent to this paradigm. I would like to thank my supervisor, Professor Stéphane Somé, for his continuous support throughout the different stages of this research effort, and for his patience while reading the various drafts. Thank you. I would also like to thank the Professor Gilbert Arbez, for the sound advice on pursuing graduate studies, and ultimately for making this opportunity a reality. I owe special thanks to the DLPS team at IBM Canada, especially Gordon McDonald, Jeff Jobb, Glynn Kneebone, and Eric Teutsch; for providing me with excellent learning experiences and numerous research opportunities, as part of the MITACS Accelerate Program. Je voudrais remercier mes parents, Guy Wilson et Lucie Dupelle, ainsi que mon frère Jeff, pour avoir cru en moi et m'avoir soutenu malgrés les multiples délais. Je vous dédie cette monographie, car sans vous elle n'aurait pas été possible. Me gustaría dar las gracias a mi novia por todo el apoyo y la comprensión. Estoy eternamente agradecido, Nina.

Virtual Environments (MMVEs) are rapidly expanding both in the number of users and complexity of interactions. Their needs of computational resources offer new challenges for the computer scientists. In this paper we present some ideas on the implementation of a Massive Simulation Environments, a particular MMVE, distributed over a Peer-to-Peer infrastructure. We analyze some of the problems related to the workload balancing on such distributed environments. In particular we discuss an hybrid Peer-to-Peer architecture in order to provide an efficient load balancing strategy. By some assumptions on temporal and spatial coherence, we use a predictor component which exploits previous phase workload as an estimate for next phase workload for load balancing purposes.

Recently, a new environment for high performance peer-to-peer distributed computing was proposed. This environment, named P2PDC, addresses stable or volatile systems communicating in a decentralized manner using the self-adaptive protocol P2PSAP. P2PDC is devoted to task parallel applications like numerical simulation problems or optimization problems solved via parallel or distributed iterative algorithms. For distributed applications meant to run with P2PDC, a performance prediction tool named dPerf was proposed. dPerf combines static and dynamic analysis with trace-based simulation to provide scientist with information about the execution of their large scale numerical simulation applications. dPerf addresses real parallel and distributed numerical simulation and optimisation applications written in C, C++ or Fortran for P2PDC. This paper introduces an enhancement of the dPerf tool which provides scalable performance prediction results. Scaling is done with respect to (i) network configuration and (ii) number of peers. Scaling predictions based on network configuration is achieved through trace-based simulation, where various architectures can be studied. Scaling predictions based on the number of peers implies analyzing the communication topology and modifying trace files prior to simulation.