This article presents a data distribution algorithm for data storage in a P2P storage system, named Us. One of the Us aims is data durability. For scalability, data are distributing on thin peers using the well known Rabin dispersal technique. Unlike other systems such as OceanStore, where data are distributed on server peers, data are distributed on end user peers. In Us, when a peer fails, a reconstruction process rebuilds lost data with help from others peers. In a previous works, we showed that for data durability such system has to face a continuous large number of reconstructions to insure data durability. To minimize end user traffic due to the reconstruction process, distribution strategies must take into account a new measure: the maximum disturbance cost of a peer during the reconstruction process. The disturbance cost is indicated by the number of data communications which are requested from a single peer for rebuilding lost data. The main goal of this article is to define algorithm able to dilute the reconstruction process in the system. We show that this problem is similar to an open mathematical problem. Hence a algorithm is defined in order to distribute data and minimize the maximum disturbance cost for each peer. Finally, toe show that our distribution algorithm is close to the non-constructive theoretical optimal distribution.

This article presents a dynamic data distribution method for data storage in a P2P system. In our system (named Us), peers are arranged in groups called metapeers to deal with account failure correlation. To minimize end user traffic according to the reconstruction process, distribution must take into account a new measure: the maximum disturbance cost of a peer. In a previous work, we defined a static distribution scheme which minimizes this reconstruction cost derived from affine plan theory. In this paper, we extend this distribution scheme to deal with the dynamic behaviour of peer to peer systems

Grand challenge applications have to process large amounts of data, and then require high performance IO systems. Cluster computing is a good alternative to proprietary system for building cost effective IO intensive platform: some cluster architectures won sorting benchmark (MinuteSort, Datamation)! Recent advances in IO component technologies (disk, controller and network) let us expect higher IO performance for data intensive applications on cluster. The counterpart of this evolution is that much stress is put on the different buses (memory, IO) of each node which cannot be scaled. In this paper we investigate a strategy we called READ/sup 2/ (Remote Efficient Access to Distant Device) to reduce this stress. With READ/sup 2/ any cluster node accesses directly to remote disk: the remote processor and the remote memory are removed from the control and data path: Inputs/Outputs don't interfere with the host processor and the host memory activity. With READ/sup 2/ strategy, a cluster can be considered as a shared disk architecture instead of a shared nothing one. This papers describes an implementation of READ/sup 2/ on Myrinet Networks. First experimental results show IO performance improvement.

Several pieces of grand challenge application software in biology or physics process large datasets which are stored on disks. These applications require high performance IO systems. Cluster computing is a good approach to build an IO intensive platform. In this paper we investigate a technique we called READ/sup 2/ which stands for Remote Efficient Access to Distant Device. The aim of this technique is to reduce the stress which is put on the buses of the cluster's node during the execution of IO streaming applications using parallel IO. In READ/sup 2/ any cluster's node directly accesses a remote disk of a distant node: the distant processor and the distant memory are removed from the control and data path. With this technique, a cluster can be considered as a shared disk architecture instead of a shared nothing one, and may inherit works from the SAN community. This paper presents what the architectural benefit of READ/sup 2/ are, i.e. a better use of IO and memory buses which eventually improve the IO scalability of a cluster and the performance of streaming applications.

This article is devoted to the run-time redistribution of one-dimensional arrays that are distributed in a block-cyclic fashion over a processor grid. While previous studies have concentrated on efficiently generating the communication messages to be exchanged by the processors involved in the redistribution, we focus on the scheduling of those messages: how to organize the message exchanges into "structured" communication steps that minimize contention. We build upon results of Walker and Otto, who solved a particular instance of the problem, and we derive an optimal scheduling for the most general case, namely, moving from a CYCLIC(r) distribution on a P-processor grid to a CYCLIC(s) distribution on a Q-processor grid, for arbitrary values of the redistribution parameters P, Q, r, and s.