This paper describes the performance benefits attained using enhanced network interfaces to achieve low latency communication.

(i.e. NIC and DMA latencies, etc) and the iteration space size and shape.

Abstract. This paper proposes a novel approach for the parallel execution of tiled Iteration Spaces onto a cluster of SMP PC nodes. Each SMP node has multiple CPUs and a single memory mapped PCI-SCI Network Interface Card. We apply a hyperplane-based grouping transformation to the tiled space, so as to group together independent neighboring tiles and assign them to the same SMP node. In this way, intranode (intragroup) communication is annihilated. Groups are atomically executed inside each node. Nodes exchange data between successive group computations. We schedule groups much more efficiently by exploiting the inherent overlapping between communication and computation phases among successive atomic group executions. The applied non-blocking schedule resembles a pipelined datapath, where group computation phases are overlapped with communication ones, instead of being interleaved with them. Our experimental results illustrate that the proposed method outperforms previous approaches involving blocking communication or conventional grouping schemes.

Abstract-The global inter-networking infrastructure that has become essential for contemporary day-to-day computing and communication tasks, has also enabled the deployment of several large-scale data sharing overlays. Communities collaboratively aggregate and distribute file and storage resources either in the controlled environment of the Grid, or hidden under the anonymity cloak created by peer-to-peer protocols. Both designs exhibit unique properties and characteristics: Peer-to-peer algorithms address the formation of vast, heterogeneous and dynamic sharing networks, while Grids focus on policy enforcement and accounting features. A distributed data management facility that will assimilate respective practices has been envisioned by numerous related research initiatives, especially when there is a need to incorporate disperse resources in large pools, without relinquishing participants of their respective rights. In this paper, we describe the Distributed File Services (DFS) architecture -a peer-to-peer service overlay, which allows distinct administrative entities to form arbitrary file distribution relationships. Each DFS peer can be uniquely authenticated and maintains direct control of its own namespace and storage assets by defining corresponding authorization directives and policies. The peerto-peer nature of the system allows for scalable deployment and resource allocation, either in a stand-alone scenario or in the Grid context. Moreover, we introduce the notion of a "web of files", as a non-hierarchical, global-scale namespace of distributed data collections and elaborate on a prototype implementation that features novel semantics for integrating our architectural principles and concepts into the operating system level.

a b s t r a c tIn this paper, we present and evaluate a protocol that enables fast and accurate range-query execution in Distributed Hash Tables (DHTs). Range queries are of particular importance when the network is populated with groups or collections of data items, whose respective identifiers are generated in a way that encodes semantic relationships into key distances. Contrary to related work in the same direction, our proposed query engine is aware of data replicas at the DHT level and by grouping related nodes into replica neighborhoods, resolves queries with the minimum amount of messaging overhead. Moreover, we suggest pairing respective operations with the core DHT routing mechanics, which allows for reusing existing management and monitoring structures and automatically adapting the query path to the dynamic characteristics of the overlay. We also present an application scenario and the respective deployment details of a prototype implementation in the context of the Gredia project.