In this paper, we introduce a majority quorum system dedicated to p-m-n general threshold schemes where p, n and m are respectively the minimal number of chunks that provide some information (but not necessarily all) on the original data, the total number of nodes in which the chunks of an object are stored and the minimal number of nodes needed to retrieve the original data using this protocol. In other words, less than p chunks reveal absolutely no information about the original data and less than m chunks can't reconstruct the original data. The p-m-n general threshold schemes optimize the usage of storage resources by reducing the total size of data to write and ensure fault-tolerance up to (n -- m) nodes failure. With such a data distribution, a specific value of m can be set to have a good trade off between resources utilization and fault-tolerance. The only drawback of such schemes is the lack of any consistency protocol. If fact, consistency protocols like classical majority quorum are based on full replication. To successfully read or write a data using the majority quorum protocol, an absolute majority of replicas must be read / written correctly. This condition ensures that any read and write operations will contain at least one common replica, which guarantees their consistency. However, when a threshold scheme is used, an adaptation is needed. In fact, classical majority quorum protocol can no longer ensure that m chunks will have the latest version when [n/2] + 1

One of the key challenges related to the grid environment is to efficiently use the storage resources of compute nodes. Therefore, we have developed the ViSaGe solution which features VisageFS - a POSIX-compliant wide-area file system, VCCC - a distributed lock mechanism, and VRT - a storage virtualization layer. In this paper, we present the VisageFStranslucentdirectoryproperty which enables a dynamic, transparent and efficient use of local storage by means of hierarchical Posix consistencies. Nodes that access a translucent enabled shared-directory will see different contents according to the site or node-level POSIX consistency. Thereafter, one of the various reconciliation modes re-enables the grid-level POSIX consistency. Our "nevermind'' mode does it inconstanttime.

A great number of scientific applications that access to a large amount of data have emerged last few years. For example in high energy physics, the volume of data will be measured in petabytes in the next years. Grids seek to aggregate geographically distributed resources for the need of these applications. It results that scientists have to deal with many problems to manage available data over the grid. Since grid resources are interconnected by WAN, the main speculation is that data transfers over the grid are bounded by network latency and bandwidth. Here, we point out some storage device issues which in many cases invalidate this assumption. Implicitly, we give a new direction that will lead to a better utilization of available resources on a grid

Data replication permits a better network bandwidth utilization and minimizes the effect of latency in large-scale systems such as computing grids. However, the cost of maintaining the data consistent between replicas may become difficult if the read/write system has to ensure sequential consistency. In this paper, we limit the overhead due to the data consistency protocols by introducing a new dynamic quorum protocol called the elementary permutation protocol. This protocol permits the dynamic reconfiguration of a tree-structured coterie (Agrawal and Abbadi, 1991) in function of the load of the machines that possess the data replicas. It applies a tree transformation in order to obtain a new less loaded coterie. This permutation is based on the load information of a small group of machines possessing the copies. The implementation and the evaluation of our algorithm have been based on the existing atomic read/write service of Lynch and Shvartsman (1997). We demonstrate that the elementary permutation ameliorates the system's throughput up to 50% in the best case. The results of our simulation show that the tree reconfiguration based on the elementary permutation is more efficient for a relatively small number of copies

Complex scientific problems involving large volumes of data need a huge computing power.Memory bandwidth remains a key issue in high performance systems.This paper presents an originrd method of memory organization, based on serial multiport memory components, which allows simultaneous access to all ports without causing either cortfhct between them or suspension.The resulting process for information exchange gives a cost-effective realization of a data memory for vector processors.The memory bandwidth can be increased considerably in a modular manner, without practical implementation constraints.The information exchange on the serial links guarantees a continuous data flow and the reaching of a maximum yield of the pipeline, thus avoiding the step-by-step feeding of the pipeline operator.

A set of ultra high throughput (more than one Gigabits per second) serial links used as processor-memory network can lead to the starting up of a shared-memory massively parallel multiprocessor. The bandwidth of the network is far beyond values found in present shared-memory multiprocessor networks. To feed this network the memory must be serially multiported. Such a multiprocessor can actually be build with current technologies.

The design of M3S, an academic shared-memory multiprocessor, is described. The memory is divided into modules. Each processor (through its cache) has an access to each memory module only through a high-throughput private serial link. Each memory module has several serial ports that are connected, in parallel, to the memory. The data coherency is maintained by a hardware directory-based scheme. The interconnection network has no bottleneck since each processor has its private path to each memory module. The high bit rate of the serial links is the most important technical problem for the design of this prototype. Synchronous solutions have been chosen in the prototype because of their greater simplicity. The data rate on the serial links is 800 Mb/s. Choices made in order to realize this project with one memory module and sixteen processor modules are explained.>