Catálogo de publicaciones - libros

The monitoring of grid environments helps administrators and users keep track of the availability and loading of resources, and the management of resources is dependent on the monitoring of information data. There is not an efficient and consistent monitoring mechanism to the manipulation of devices, resources and services in Grid computing. We propose a novel monitoring framework used to gather and retrieve monitoring information of Grid environments. The monitoring system RMCS integrates and extends the existing monitoring system using service-oriented mechanism and the common information model CIM. The RMCS defines a hierarchical structure of monitoring resources, and customizes the monitoring parameters and the display way. The adoption of CIM-based monitoring service enables compatible with other grid services such as grid portal, transaction or resource management and charging. The investigation shows that this monitoring approach provides the scalable monitoring capabilities, enables to exchange information in an unrestricted and flexible way, and improves grid performance and utilization.

In this paper, we proposed distributed and adaptive grid workflow net model. Which applies the Coloured Petri net as the formalism to describe grid process, and proposed the formal method for grid services to composite the gridflows. And also proposed the multi-agent based implementation for gridflow net model.

Many emerging group oriented and collaborative applications such as audio/video conferences use the peer-to-peer (P2P) paradigm. Confidentiality is an often demanded feature for such applications, e.g. in business meetings, to provide group privacy. How to build a secure P2P video conference system is still an open issue. In this paper several possible solutions are discussed. We present a security architecture used for P2P video conferences that ensures confidential talks in an enterprise environment whose branches might be geographically dispersed.

Peer-to-Peer (P2P) architectures are very prevalent in today’s Internet. Lots of P2P file sharing systems using Gnutella protocol emerge out and draw attractions of millions of people. The “flooding” search mechanism of Gnutella makes it easy to be deployed, but also spawns numerous messages which leads to serious scalability problems. However, the locality discovered in both user’s share files and queries, enables us to use query-caching to shorten the search length and reduce the messages traffic. This paper makes an extensive study of query-caching in P2P systems and proposes an adaptive query-caching mechanism to manage the cached query reply messages according to the heterogeneity of the uptime of different peers. Along with several other techniques we proposed, our approach achieves a 30% reduction of average search length and a 61% reduction of query message traffic comparing with the previous query-caching mechanisms in the simulation, which indicates that our approach makes Gnutella more scalable.

This paper presents the design and deployment of a locality-aware overlay multicast protocol called . The key idea of is to use the geometrical information of end hosts to construct the locality-aware overlay data delivery tree such that nearby users in the underlying network can be organized into nearby subtrees. The prototype of has been widely used in CERNET. Logging traces obtained from broadcasting 2004 Athens Olympic Games over 16 days have shown that the performance of , such as end-to-end delay and absolute data delivery delay, significantly outperforms that of randomly constructed overlay multicast.

With the development of high-speed backbone network, more and more traffic load is pushed to the Internet end system. The satisfactory execution of common business applications depends on the efficient performance of web server. In this paper, we propose a pipeline multi-thread kernel web server open KETA which divides the processing of a request into several independent phases. This architecture reduces parallelism granularity and achieves inner-request parallelism to enhance its processing capability. Furthermore, a thread allocation model is used to manage threads effectively in this special architecture. This model can adjust the thread allocation based on the server load and the work character of each phase so that the thread resource of web server can be utilized properly. Experimental result shows the capability of this web server and the effectiveness of the thread allocation model.

Service composition has been recognized as a flexible way for resource sharing and application integration. Quality of service (QoS) is an important issue for composite services. In this paper, we address the issue of component services selection to ensures their composition satisfy given QoS constraints. We propose the concept of reduction tree as a general scheme to aggregate multi-dimensional quality. And then a heuristic algorithm MCSC_HEU is presented to find execution plans satisfying multiple QoS constrains, with the main idea of evaluating partial plans by a heuristic function during the course of reduction. The time complexity of MCSC_HEU is of polynomial level. Extensive evaluations show that MCSC_HEU succeeds in finding feasible plans with very high probability but demands much less time than exhausting search. So it is an efficient solution for QoS aware service selection with multiple constrains.

This paper focuses on the Cyclops64 computer architecture and presents an analytical model and performance simulation results for the preloading and loop unrolling approaches to optimize the performance of SVD (Singular Value Decomposition) benchmark. A performance model for dissecting the total execution cycles is presented. The data preloading using “memcpy” or hand optimized “inline” assembly code, and the loop unrolling approach are implemented and compared with each other in terms of the total number of memory access cycles. The key idea is to preload data from offchip to onchip memory and store the data back after the computation. These approaches can reduce the total memory access cycles and can thus improve the benchmark performance significantly.

In this paper, we propose a novel scheme, named as TCP-ABC, which replicates the server side TCP connections among multiple server nodes of a cluster. By guaranteeing atomic request delivery, and consensus on responses, this scheme provides the legacy server applications running on the server nodes with multiple active backups in a transparent fashion. By failing the connections over healthy units, the scheme enhances the service and data availability of the cluster. By conducting experiments on the prototype system of a cluster up to four nodes, we find TCP-ABC results in small performance lost while greatly enhances the service and data availability.

In this paper we introduced a parallel file system based on the spatial information object storage, the PIPFS system. PIPFS is a special-purpose parallel file system which designed in view of the remote sensing image processing. It uses the server/client pattern and bases on the metadata mechanism. It simultaneously accesses disks on several nodes for application I/O operations, which improves the efficiency of the operation on large scale data. A high performance is shown on high-data-complexity application, such as remote sensing image processing.