Abstract not found

Hidden variables, evolving over time, appear in multiple settings, where it is valuable to recover them, typically from observed sums. Our driving application is 'network tomography', where we need to estimate the origin-destination (OD) traffic flows to determine, e.g., who is communicating with whom in a local area network. This information allows network engineers and managers to solve problems in design, routing, configuration debugging, monitoring and pricing. Unfortunately the direct measurement of the OD traffic is usually difficult, or even impossible; instead, we can easily measure the loads on every link, that is, sums of desirable OD flows. In this

Abstract — In a grid-computing environment, resource selection and scheduling depend on the network topology connecting the computation nodes. This paper presents a method to hierarchically group compute nodes distributed across the internet into logical clusters, and determine the relative location of the clusters. At inter-domain level, distance from landmarks (a small group of distributed reference nodes) is the basis for converting the location of nodes inside a complex network structure onto a simple geometric space. The position of compute nodes in this geometric space is the basis for partitioning nodes into clusters. For compute nodes within an administrative domain, minimum RTT is used as the metric to partition nodes into clusters. This approach leads to an efficient, scalable and portable method of clustering grid nodes and building a distance map among clusters. We demonstrate the system for automatic clustering by applying it to computation nodes distributed across five universities in Texas. I.

We present an MPI topology discovery tool for homogeneous Ethernet switched clusters. Unlike existing Ethernet topology discovery methods that rely on Simple Network Management Protocol (SNMP) queries to obtain the topology information, our tool infers the topology from end-toend measurements. The tool works on clusters connected by managed and/or unmanaged Ethernet switches, and does not require any special privilege. We discuss the theoretical foundation of the tool, present the algorithms used, and report our evaluation of the tool. 1

Abstract — Information about the topology and link-level characteristics of a network is critical for many applications including network diagnostics and management. However, this information is not always directly accessible; subnetworks may not cooperate in releasing information and widespread local measurement can be prohibitively expensive. Network tomographic techniques obviate the need for network cooperation, but the majority assume probing from a single source, which imposes scalability limitations because sampling traffic is concentrated on network links close to the source. We describe a multiple source, end-toend sampling architecture that uses coordinated transmission of carefully engineered multi-packet probes to jointly infer logical topology and estimate link-level performance characteristics. We commence by demonstrating that the general multiple source, multiple destination tomography problem can be formally reduced to the two source, two destination case, allowing the immediate generalization of any sampling techniques developed for the simpler, smaller scenario. We then describe a method for testing whether links are shared in the topologies perceived by individual sources, and describe how to fuse the measurements in the shared case to generate more accurate estimates of the link-level performance statistics. Index Terms — Internet tomography, end-to-end measurements, active probing, topology discovery, loss rate estimation.

Abstract. This study deals with the topology discovery for the capacitated minimum spanning tree network. The problem is composed of finding the best way to link nodes to a source node and, in graph-theoretical terms, it is to determine a minimal spanning tree with a capacity constraint. In this paper, a heuristic algorithm with two phases is presented. Computational complexity analysis and simulation confirm that our algorithm produces better results than the previous other algorithms in short running time. The algorithm can be applied to find the least cost multicast trees in the local computer network. 1

Tunnels are widely used to improve security and to expand networks without having to deploy native infrastructure, and play an important role in the migration to IPv6. In this paper we introduce a number of techniques to detect, and collect information about, IPv6-in-IPv4 tunnels. We also show how, once a tunnel has been discovered, it can be used as a “vantage point ” to launch third-party tunnel-discovery explorations, scaling up the discovery process. We describe the Tunneltrace tool which implements the proposed techniques, and validate them by means of a wide experimentation on the 6bone tunneled network, on the Italian Academic and Research network, and through the test boxes deployed worldwide by the RIPE NCC as part of the Test Traffic Measurements Service. We assess to what extent 6bone registry information is coherent with the actual network topology, and we provide the first experimental results on the current distribution of IPv6-in-IPv4 tunnels in the Internet, showing that tunnels are very common: even the “native ” networks we tested reach more than 60 % of all IPv6 prefixes through tunnels.

Abstract. The concept of using Internet edge nodes for High Performance Computing (HPC) applications has gained acceptance in recent times. Many of these HPC applications also have large I/O requirements. Consequently, an edge node file system that efficiently manages the large number of files involved can assist in improving application performance significantly. In this paper, we discuss the design of a Distributed File System (DFS) specialized for HPC applications that exploits the storage, computation and communication capabilities of Internet edge nodes and empowers users with the autonomy to flex file management strategies to suit their needs. Key words: distributed file systems, high performance computing, peer-to-peer systems, scalability

Abstract not found

Abstract — ICT systems are made of software, middleware and hardware components and are usually distributed over a network. Middleware and hardware components are combined on what is usually designated an "IT Infrastructure (ITI)". It is upon that ITI that application software is run. The topology of that ITI poses constraints on software algorithms, data structures and software configuration, due to concerns such as fault tolerance, latency or synchronization. In large ITIs that topology evolves constantly. Therefore, when faced to the challenge of maintaining a legacy ICT system, it is important to use reengineering techniques to discover the ITI topology. We propose a reengineering technique to discover the topology of a distributed IT infrastructure, based on a multinomial logistic regression model and a set of topology stereotypes. To demonstrate the feasibility of the approach we applied the model to several organizations with distributed ITIs and, among other aspects, we found that the most recurrent stereotypes are the centralized and backbone ones.