Catálogo de publicaciones - libros

This paper addresses the optimization problems of placing servers in the presence of competition. We place a set of on a graph to compete with the set of . Our objective is to find the placement that maximizes the benefit, which is defined as the profits from the requests made to the extra servers despite the competition, minus the cost of constructing those extra servers.

We propose an (||) time dynamic programming algorithm to find the optimal placement of extra servers that maximizes the benefit in a tree with || nodes. We also propose an (||) time dynamic programming algorithm for finding the optimal placement of extra servers that maximizes the benefit, without any constraint on the number of extra servers. For general connected graphs, we prove that the optimization problems are NP-complete. As a result, we present a greedy heuristic for the problems. Experiment results indicate that the greedy heuristic achieves good results, even when compared with the upper bounds found by a linear programming algorithm. The greedy heuristic yields performances within 15% of the upper bound in the worst case, and within 2% of the same theoretical upper bound on average.

Semantic service discovery improves the performance of service matching, due to using ontology and logical reasoning. However, in open distributed computing environment available mechanisms for semantic service discovery face new challenges: increasing scale of systems, multiple coexistent ontologies and so on. Aiming to these problems, a semantic service discovery mechanism based on ontology community, SSD_OC, is proposed in this paper. Multiple coexistent ontologies are supported by SSD_OC and bridging axioms between different ontologies enable users to match services across ontologies. Experiment results show that SSD_OC is scalable and outperform other systems in term of F-Measure.

In this paper, we present a new extension and reinterpretation of one of the most successful models of awareness in Computer Supported Cooperative Work (CSCW), called the Spatial Model of Interaction (SMI), which manages awareness of interaction through a set of key concepts, to provide task delivery in collaborative distributed systems. This model also applies some theoretical principles and theories of multi-agents systems to create a collaborative and cooperative environment that can be able to provide an autonomous, efficient and independent management of the amount of resources available in a cluster. This model has been implemented in a cluster based on a multi-agent architecture. Some results are presented with the aim of emphasizing the performance speedup of the system using the Collaborative Awareness Model for Task-Balancing-Delivery (CAMT).

In this paper, we propose an improved model for predicting HPL (High performance Linpack) performance. In order to accurately predict the maximal LINPACK performance we first divide the performance model into two parts: computational cost and message passing overhead. In the message passing overhead, we adopt Xu and Hwang’s broadcast model instead of the point-to-point message passing model. HPL performance prediction is a multi-variables problem. In this proposed model we improved the existing model by introducing a weighting function to account for many effects such that the proposed model could more accurately predict the maximal LINPACK performance . This improvement in prediction accuracy has been verified on a variety of architectures, including IA64 and IA32 CPUs in a Myrinet-based environment, as well as in Quadrics, Gigabits Ethernet and other network environments. Our improved model can help cluster users in estimating the maximal HPL performance of their systems.

Limited computing resources may often cause poor performance and quality. To overcome these limitations, we introduce the idea of ad hoc systems, which may break the resource limitation and give mobile devices more potential usage. That is, several resource-limited devices may be combined as an ad hoc system to complete a complex computing task. We illustrate how the adaptive software framework, , may realize ad hoc systems by automatically distribute software to appropriate devices via the assembly process. We discuss the problem that ad hoc systems may be unstable under mobile computing environments since the participating devices may leave the ad hoc systems at their will. We also propose the reassembly process for this instability problem; i.e., assembly process will be re-invoked upon environmental changes. To further reduce the performance impact of reassembly, two approaches, partial reassembly and caching, are described. Our experimental results show that the caching improves performance by a factor of 7 ~40.

This paper focuses on two problems related to QoS-aware I/O server placement in hierarchical Grid environments. Given a hierarchical network with requests from clients, the network latencies of links, constraints on servers’ capabilities and the service quality requirement, the solution to the attempts to place the minimum number of servers that meet both the constrains on servers’ capabilities and the service quality requirement. As our model considers both the different capabilities of servers and the network latencies, it is more general than similar works in the literatures. Instead of using a heuristic approach, we propose an optimal algorithm based on dynamic programming to solve the problem. We also consider the , which tries to place a given number of servers appropriately so that the maximum expected response time is minimized. We prove that an optimal server placement can be achieved by combining the dynamic programming algorithm with a binary search on the service quality requirement. The simulation results clearly show the improvement in the number of servers and the maximum expected response time.

In this paper, we present a novel self-healing engine for autonomic network management. A light weight Self Management Frame (SMF) performs monitoring and optimization functions autonomously and the other self management functions, driven by context, are invoked on demand from the server. The policies are maintained to calculate the trust factor for network entities and those trust factors will be used at the later stages of our project to enforce resource utilization policies. The plug-ins, residing at the server, are used to perform the on-demand management functions not performed by SMF at client side. A Simple Network Management Protocol (SNMP) based monitoring agent is applied that also triggers the local management entities and passes the exceptions to the server which determines the appropriate plug-in. Considering the amount of resources being put into current day management functions and contemporary autonomic management architectures our findings show improvement in certain areas that can go a long way to improve the network performance and resilience.

The discovery and selection of needed resources, taking into account optimization criteria, local policies, computing and storage availability, resource reservations, and grid dynamics, is a technological challenge in the emerging technology of grid computing.

The Condor Project’s ClassAd language is commonly adopted as a “” for describing grid resources, but Condor itself does not make extensive use of Web Services. In contrast, the strongly service-oriented Globus Toolkit is implemented using the web services resource framework, and offers basic services for job submission, data replica and location, reliable file transfers and resource indexing, but does not provide a resource broker and matchmaking service.

In this paper we describe the development of a Resource Broker Service based on the Web Services technology offered by the Globus Toolkit version 4 (GT4). We implement a fully configurable and customizable matchmaking algorithm within a framework that allows users to direct complex queries to the GT4 index service and thus discover any published resource. The matchmaking algorithm supports both the native simple query form and the Condor ClassAd notation. We achieve this flexibility via a matchmaking API java class framework implemented on the extensible GT4 index service, which maps queries over ClassAds in a customizable fashion.

We show an example of the proposed grid application, namely an on demand weather and marine forecasting system. This system implements a Job Flow Scheduler and a Job Flow Description Language in order to access and exploit shared and distributed observations, model software, and 2D/3D graphical rendering resources. The system combines GT4 components and our Job Flow Scheduler and Resource Broker services to provide a fully grid-aware system.

The trustworthiness between anonymous web service client and provider influences service stability and collaboration. Trustworthiness includes some aspects such as service’s security, controllability and survivability. So, a definition of trustworthiness for web service is given in the paper, and then a web service trustworthiness evaluation and management model is brought forth inspired by human small-world network. The model consists of three web service federations: WSRRC, APAEAS and AWSORT. WSRRC is a Web Service Resource Register Center, which is established by UDDI protocol. APAEAS is an Area Proxy Authentication Evaluating Autonomy System, which collects some authentication information of web service clients, accepts clients’ special requirement and feedbacks service’s trustworthiness values to AWSORT. AWSORT is an Area WS Resource Organizing Tree, which organizes and manages web service resources; records web service trustworthiness values, keeps web service state, assigning web service. The model establishes a trustworthy environment for anonymous web service clients and providers. Furthermore, some detailed evaluating parameters about service trustworthiness and quality is discussed and some service management algorithms are proposed in the paper. The simulation results show that model is feasible for semantic grid integration and establishment for virtual organization.

A grid has to provide strong incentive for participating sites to join and stay in it. Participating sites are concerned with the performance improvement brought by the gird for the jobs of their own local user communities. Feasible and effective load sharing is key to fulfilling such a concern. This paper explores the load-sharing policies concerning feasibility and heterogeneity on computational grids. Several job scheduling and processor allocation policies are proposed and evaluated through a series of simulations using workloads derived from publicly available trace data. The simulation results indicate that the proposed job scheduling and processor allocation policies are feasible and effective in achieving performance improvement on a heterogeneous computational grid.