INDEX
| Raweb 2003 / Project-Team : reso
|
HELP INDEX |
|
Key words: network service, cost estimation, replica optimization.
In a Data Grid, replicas are located at several different storage locations with a large range of possible current network throughput and latency. It is important to select replicas based on their minimal access latency. Although replica access optimization does not only depend on the network link and its capacity (the load and the latency of data servers have to be taken into account too), we have examined how an aggregate knowledge of the network behavior may have an important impact on the replica access optimization step.
In the last few years, dedicated Grid network monitoring systems have been developed and deployed within Grid environments [85] , [84] . To provide the users with an abstract and homogeneous view of the complex set of interconnected resources, we designed and developed a performance measurement system that is characterized by simple and relevant metrics of a grid network cloud [28] , [27] . However, to optimize the application performance, a Grid middleware component requires aggregate and simple estimations of transfer costs between defined end hosts. We examine how high level Network Cost Estimation Functions (NCEF) can be computed and used in a Grid environment for network-based replica optimization [83] . In particular, we study how an estimation of the end to end transfer delay of a certain amount of data can be easily derived from the raw network performances measurements. A flexible and open Grid Network Cost Estimation Service (NCES) that permits Grid resource management services to use network monitoring data in a very simple fashion has been developed. We demonstrate that provided approximations are valuable in certain cases, are necessary to find a tradeoff between accuracy, efficiency and scalability, and to define an extensible set of functions within the framework of an open service.
Key words: Grid, active networks.
We have studied the benefits of programmable and active networks as an alternative solution for dynamic deployment of networks services adapted to Grid infrastructure. This proposition is called ``Active Grid''. Preliminary works based on Tamanoir proposed collaborative usage of high performance execution environment with Grid middleware and applications [70] [64] [29] [30] .
Key words: cryptographic Identifiers, distributed security, decentralized security, applications, grid.
A new framework that will eventually allow to seamlessly secure any distributed application was described and partially implemented in [26] and [16] . This framework rests on the use of CBIDs by each entity combined with the use of SPKI authorization certificates, thus allowing a given CBID to delegate rights to another CBID. Amongst others things, this allows to secure the distributed and shared remote storage protocol Internet Backplane Protocol.
Key words: Identifiers, distributed security, decentralized security, identification, localization.
The implementation of an infrastructure of cryptographic identifiers in the network layer shows well the fundamental utility of such an infrastructure dice the network layer because it allows two nodes previously unknown from each other to communicate in a protected way on level IP This implies two other nodes acting as footbridges of IPsec safety, which will have to be discovered mutually and to exchange certificates of delegation proving that they are authorized by the two extremities of the data flow to act of the kind. Given the experimental and applied nature of this work, it could only profit from a use in real conditions, showing in this way its applicability and its practicality, it is judicious to standardize the stable components ``as far as possible'' of the aforesaid work, so that their use can spread in a way increased in the communities likely to use them. To this end, a work of standardization was started within the IETF around techniques of CGA/CBID, as described in [50] , [52] and [51] .
Key words: .
As part of the RNTL e-Toile project, the communication library, Madeleine [53] , designed by the R. Namyst team at Bordeaux has been ported over Globus. This gives the opportunity to use the native Madeleine API as well as the MPI API. The port gives to the applications an API whose model is fully connected without having to establish the connections between the nodes nor having to know the underlying network (TCP, cluster, grid). Madeleine gives the possibility to use different bufferisation modes for transferring the data. More precisely as part of the RNTL e-Toile project, Madeleine transparency has been generalized to a multi-site deployment with the e-Toile authentification and security framework.
Key words: Reliable multicast, programmable networks, computational grids.
With a logical view closer to a distributed operating system than a pure communication infrastructure, one might consider extending for computational grids the basic functionalities found in the commodity Internet's network infrastructure. Our work on multicast support for the grid is based on the possibility to easily (at least more easily than on the Internet) add processing elements (active routers) in the network infrastructure of a grid. These ideas and motivations are described in [78] , [75] and [20] for multicast communications.
A first prototype of DyRAM has been developed by a master student, J. Mazuy, in 2002. The performance results have been published in [72] . This prototype has been improved and extended in the VTHD++ and E-Toile projects. We are implementing an ftp-like tool for data and code transfers on a computational grid. This work has lead to seminars and a demonstration at IPDPS 2003 (ACI GRID booth) and within the VTHD++ and E-Toile projects.
|
|
|