Catálogo de publicaciones - libros

In this paper we present PLF (Promotion of Long Flows). PLF tries to promote web traffic using the Token Bucket Model in a DiffServ framework. This algorithm preserves the high priority for short flows but tries to allocate some long flows in the high priority class level, in order to improve some performance parameters of the long ones. Finally, we present PLFwp (Promotion of Long Flows with Penalization) in order to detect (and cancel) the effect of these extremely long flows over the global performance. We analyze the results for packet loss and web transmission latency.

Converged network seamlessly integrates different communications media such as data, voice and multimedia on a single platform. It refers to convergence both types of network and technologies as well as convergence between the different layers of network architecture. In this paper, we examine a priority-based queuing model and perform the mathematical analysis of different media calls processing in converged network environment. We use for this purpose a queuing system model M3/G3/1/NPRP in order to process effectively input jobs/requests (or packets). Tasks within this queuing system get a higher priority if they are handling a real-time event. We present in our paper mathematical results of the expected response and waiting time, and build hypothetical diagrams for the further practical usage in real-time system. A modeling method developed in this paper will be used for the fast configuration and testing of new converged network applications and services.

We have witnessed rapid growth of mobile Internet capabilities in the mobile handsets. The penetration of the user life increases the need to establish a methodology to capture the user behavior patterns in order to improve satisfaction of the mobile web users. In order to cope with the design evaluation needs, the author proposes a long-term command transition probability analysis on mobile clickstreams. The observation showed that the first command ratio did not depend on the menu order, but that the ratio to terminate the command sequence depended on the order. This gives a positive outcome from the command transition analysis in the mobile clickstream.

In mobile ad hoc networks, most of on demand routing protocols such as DSR and AODV do not deal with traffic load during the route discovery procedure. To solve the congestion and achieve load balancing, many protocols have been proposed. However, existing load balancing schemes just avoid the congested route in the route discovery procedure or find an alternative route during a communication session. To solve this problem, we propose a new scheme which uses the packet bursting mechanism in congested nodes. The packet bursting, which is originally introduced in IEEE 802.11e QoS specification, is to transmit multiple packets after channel acquisition. Thus, congested node can forward buffered packets quickly and prevent bottleneck. Each node begins to transmit packets in normal mode whenever its congested status is dissolved. We also propose two threshold values to define exact overloaded status adaptively; one is interface queue length and the other is buffer occupancy time. Through a experimental simulation study, we compare our protocol with normal on demand routing protocols and show that the proposed scheme is more efficient and effective especially when network traffic is heavily loaded.

In this paper, we present the results of our analysis in which we investigated the different existing peer-to-peer systems with regard to design pattern usage. In the course of our analysis, we mainly concentrated on patterns concerned with the classification of distributed systems. In the following, the design patterns investigated are examined in more detail in their usage context and possible alternative solutions are discussed.

In any distributed application, the communication between the distributed processes/nodes of the distributed systems is essential for both reliability and efficiency matters. In this work we address this issue for distributed applications based on JXTA protocols. After a careful examination of the current version of JXTA protocols, we observed the need for improving the original JXTA protocols, such as pipe services, to ensure reliable communication between peer nodes and the discovery and presence service to increase the performance of the applications. The re-implemented protocols have been validated in practice by deploying a P2P network using nodes of PlanetLab platform and testing each of the extended protocols using this real P2P network.

We propose a peer-to-peer system that supports distributed virtual world applications. For these applications, the connections between directly neighboring peers are of the utmost importance. To minimize wide area network traffic and average latency, peers that belong to the same subnet, are grouped together, and these groups are interconnected via wide area connections. To build up and maintain this optimized peer-to-peer structure, we developed a set of protocols that efficiently handle the joining and leaving of peers as well as failure situations. Peers are aranged in the logical ring structure using a two-step discovery and join procedure. The first step uses broadcast messages to discover peers in a local subnet, followed by a local join. If no peer answers in the local subnet, a remote join is performed. With the implemented recovery procedures, our peer-to-peer system can survive multi-node failures in a local subnet as well as the failure of an entire subnet.

A challenging problem in a schema-based peer-to-peer (P2P) system is how to locate peers that are relevant with respect to a given query. In this paper, we propose a new semantic routing mechanism in the context of the SenPeer (PDMS). SenPeer is an unstructured P2P system based on an organization of peers around super-peers according to their semantic domains. Our proposal is based on the use of a distributed data structure, called expertise table, maintained by the super-peers and describing data at the neighboring peers. This table, combined with matching techniques, is the basis of a semantic overlay network. Semantic links are exploited for efficient query propagation towards peers that may have relevant data. We give a performance evaluation of our semantic query routing with respect to important criteria such as precision, recall and number of messages. The results show that our algorithm significantly outperforms a baseline algorithm without semantics.

Peer-to-Peer computing, which aggregates idle computing cycles and storage space from PCs on the Internet, is a new approach to establish a high performance computing system. In this paper, we introduce Jamjuree Cluster, a cluster middleware suite that is an integration of cluster computing and peer-to-peer computing concepts. Jamjuree Cluster creates a single system image from PCs on a peer-to-peer network, and provides a parallel programming environment, file system, and batch scheduling like in a typical cluster system.

Recently, in Japan, some farming districts established their locality as brands, and prices of agricultural products differs from their grown places. This induced some agricultural food origin forgery cases. Food traceability systems are introduced and some are now in operation to solve this problem. However, food traceability systems have vulnerabilities in their nature because they traces only artificially attached IDs. So there are possibility to forge ID and packages, and switching the vegetables in packages. So, we developed a geographical origin identification system for vegetables by using their trace element compositions. Trace element means very small quantities of elements. This system gathers trace element data of vegetables when shipping from farms, and stores them into databases located in farming districts. In case of a vegetable which has doubtful geographical origin is found in markets, their trace element compositions are measured and compared with data in databases to find its actual geographical origin. Our system judges geographical origin by whether correlation coefficient. This requires calculating correlation coefficients for identifying one and all stored data. However, this is not scalable for the number of data. In this paper, we describe a method to limit the number of data to be used to calculate correlation coefficients before calculating them, and realize scalability.