Tesis doctoral de Daniel Lázaro Iglesias
Cloud computing has recently emerged as a powerful new computing paradigm causing many companies and organizations to move their software to large data centers owned by specialized resource providers. large data centers are also used to host and support internet communities, like social networks (e.G. facebook, twitter, flickr, etc), which constitute cloud services offered directly to end users. At the same time, millions of individuals, as well as enterprises, own underutilized computers connected to the internet. While some models have appeared to use these resources for executing computational tasks, like volunteer computing and desktop grids, these usage models for surplus resources do not fit the needs of cloud consumers, individual service providers or community members. In parallel to these advances, peer-to-peer (p2p) networks have become popular among internet users and a growing field of research for developing distributed systems with desirable features like decentralization, self-organization, fault tolerance and high scalability among others. However, the p2p paradigm has been limited in practice to applications sharing disk space and network bandwidth, and sharing compute power through p2p networks has not yet become a reality. we address the lacks of cloud computing, desktop grids and peer-to-peer networks to use nondedicated resources for general purpose computing by proposing the model of contributory computing, where users contribute their resources to be used collectively. A contributory community (aggregation of contributed resources) can be used as a platform to deploy services, as an alternative to large data centers. This services make it possible to use surplus resources in a general way, instead of limiting its use to the execution batch tasks. this thesis develops the concept of contributory computing by presenting a middleware for building contributory communities and deploying services on the contributed resources. This middleware, called codes, allows resources to be put into the community and services deployed on it, and takes care of keeping the services available using the contributed resources. It is self-managed, decentralized, scalable, fault-tolerant and can deal with heterogeneous resources. Specifically, this thesis will focus on defining an architecture for the middleware and designing some of the main mechanisms that allow the creation of contributory communities. The designed mechanisms are a) a mechanism for service deployment, b) a resource discovery mechanism, and c) an availability-aware resource selection mechanism. these mechanisms are decentralized, scalable and failure tolerant. Specifically, to propose our resource selection mechanism we study resource availability in volunteer computing systems, and use a set of real system traces to develop an availability prediction method. This predictor is used in the design of the resource selection mechanism for service deployment, which can leverage availability information to provide service availability while minimizing the number of migrations. All these contributions serve to prove the technical feasibility of contributory computing. Moreover, we have implemented a prototype of the designed middleware and mechanisms, and have deployed it over planetlab, a real distributed testbed. This forms a basis to build real contributory systems.
Datos académicos de la tesis doctoral «A middleware for service deployment in contributory computing systems«
- Título de la tesis: A middleware for service deployment in contributory computing systems
- Autor: Daniel Lázaro Iglesias
- Universidad: Oberta de catalunya
- Fecha de lectura de la tesis: 06/07/2011
Dirección y tribunal
- Director de la tesis
- Joan Manuel Marques Puig
- Tribunal
- Presidente del tribunal: leandro Navarro moldes
- nuno Manuel Ribeiro preguií§a (vocal)
- (vocal)
- (vocal)