Angelo Corana

Job-resource matchmaking on Grid through two-level benchmarking

Grid environments must provide effective mechanisms able to select the most adequate resources satisfying application requirements. A description of applications and resources, grounded on a common and shared basis, is crucial to favour an effective pairing. A suitable criterion to match demand with supply is to characterize resources by means of their performance evaluated through the execution of low-level and application-specific benchmarks. We present GREEN, a distributed Matchmaker, based on a two-level benchmarking methodology. GREEN facilitates the ranking of Grid resources and the submission of jobs to the Grid, through the specification of both syntactic and performance requirements, independently of the underlying middleware and thus fostering Grid interoperability.

Resource selection and application execution in a grid: a migration experience from GT2 to GT4

The latest Globus Toolkit 4 (GT4) version, fully OGSA compliant and almost completely based on Grid Services, is expected to become the reference version of the Globus Toolkit Therefore, the necessity to migrate from previous versions, especially the widespread GT2, to GT4 is becoming a relevant issue for many Grid systems. We present a migration experience from GT2 to GT4 of a simple tool for resource discovery and selection, and execution of parallel applications in a computational Grid environment. The proposed tool provides a simple and intuitive user interface, a structured view of Grid resources, a simple mechanism for resource selection, and support for application staging and execution. We discuss some relevant differences between GT2 and GT4, particularly for the Monitoring and Discovering subsystem GT4 turns out to be more powerful and flexible. Just because we address the implementation of basic functionalities, the proposed discussion may be of general interest.

Parallel computation of the correlation dimension from a time series

Abstract A parallel algorithm is presented for computing the Correlation Dimension ( D 2 ) from a time series generated by a dynamical system, using the method of correlation integrals. Three versions are discussed: the first computes all distances between points in the phase space, whereas the second and third compute only distances less than a threshold ϵ ; the third version in particular is very powerful since it employs a box-assisted approach and linked lists for a fast search of neighboring points. The parallelization is designed for coarse-grained multiprocessor systems with distributed memory and is accomplished using a message passing model and partitioning points evenly among processors. Uniform implementation and computational analysis allow a clear comparison of the three versions. The algorithms, tested on the Transtech PARAstation multiprocessor, are well balanced, give a linear speed-up and show a good scalability. The third version is particularly suitable for fast processing of very long time series and allows the estimation of D 2 even for medium- and high-dimensional systems, where an extremely large number of points is needed. The algorithms can be adapted with few modifications to the computation of the generalized dimensions D q , and they can also be useful in other applications involving the efficient computation of distances between points in a large set. More generally, the computational framework can be used in similar problems involving long-range interactions.

Quality of Service on Grid: architectural and methodological issues

The rapid evolution of Grid Computing and the development of new middleware services make Grid platforms increasingly used not only for best effort scientific jobs but also in industrial and business applications. This has taken to the growing demand of Quality of Service (QoS) support. However, the QoS issue on Grid is quite difficult, as Grid has been originally designed without any QoS support, and it is a complex system. During the previous years some solutions have been proposed to supply QoS for specific classes of applications. This results in a focused and a heterogeneous approach, so that it is hard to evaluate its sufficiency and its robustness with respect to the large spectrum of possible Grid applications. In this context, our contribution concerns three points: first, we analyze the current approach to QoS on Grid as a relationship among QoS features, applications, and architectures; second, we evaluate the QoS requirements of two recent QoS‐demanding applications on Grid, namely Massive Multiplayer Online Games and Urgent Computing, comparing these requirements with the support provided by current QoS architectures; and third, we propose an alternative approach to QoS provision on Grid based on the definition of a dedicated QoS‐management layer to overcome the limitations of the current methodologies. Copyright

A Distributed Approach for Structured Resource Discovery on Grid

We present a distributed approach for grid resource discovery, which combines a structured view of resources (single machines, homogeneous and heterogeneous clusters) at the physical organization (PO) level with a super-peer network connecting the various POs. The proposed architecture is modular and independent of the particular grid middleware. After a general description, we present some implementation aspects which refer to the Globus Toolkit 4 as grid middleware and to the JXTA platform to set-up the super-peer network. The system is particularly suitable for discovering resources for structured parallel applications on very large grids.

Performance Analysis of Task-Based Algorithms on Heterogeneous Systems with Message Passing

We address the problem of performance analysis and prediction of a class of parallel applications on heterogeneous systems. Our attention is oriented towards workstation networks programmed using message passing libraries. Particularly, we consider a switched Ethernet-based network and we use PVM as parallel tool, adopting the master-worker model with the task farm paradigm.

Computing the correlation dimension on a network of workstations

SUMMARY We present a parallel algorithm for computing the correlation dimension (D2) from a time series generated by a dynamic system, using the method of correlation integrals, which essentially requires the computation of distances among a set of points in the state space. The parallelization is suitable for coarse-grained multiprocessor systems with distributed memory and is carried out using a virtually shared memory model. The algorithm simultaneously gives all the correlation integrals at various state space dimensions needed to estimate the D2. Two versions are discussed: the first computes all distances between points; the second computes only distances less than a fixed, and employs a boxassisted approach and linked lists for an efficient search of neighbouring points. The algorithms, coded in Fortran 77, are tested on a heterogeneous network of workstations consisting of various DEC Alphas of different powers, interconnected by Ethernet; the Network Linda parallel environment is used. A detailed analysis of performance is carried out using the generalization of speed-up and efficiency for heterogeneous systems. The algorithms are fully asynchronous and so intrinsically balanced. In almost all the situations they provide a unitary efficiency. The second version greatly reduces the computational work, thus making it possible to tackle D2 estimation even for medium and high-dimensional systems, where an extremely large number of points is involved. The algorithms can also be employed in other applicative contexts requiring the efficient computation of distances among a large set of points. The method proposed for the analysis of performance can be applied to similar problems. ©1998 John Wiley & Sons, Ltd.

PORTING REGULAR APPLICATIONS ON HETEROGENEOUS WORKSTATION NETWORKS: PERFORMANCE ANALYSIS AND MODELING

Abstract Heterogeneous networks of workstations and/or personal computers (NOW) are increasingly used as a powerful platform for the execution of parallel applications. When applications previously developed for traditional parallel machines (homogeneous and dedicated) are ported to NOWs, performance worsens owing in part to less efficient communications but more often to unbalancing. In this paper, we address the problem of the efficient porting to heterogeneous NOWs of data-parallel applications originally developed using the SPMD paradigm for homogeneous parallel systems with regular topology like ring. To achieve good performance, the computation time on the various machines composing the NOW must be as balanced as possible. This can be obtained in two ways: by using an heterogeneous data partition strategy with a single process per node, or by splitting homogeneously data among processes and assigning to each node a number of processes proportional to its computing power. The first method is however more difficult, since some modifications in the code are always needed, whereas the second approach requires very few changes. We carry out a simplified but reliable analysis, and propose a simple model able to simulate performance in the various situations. Two test cases, matrix multiplication and computation of long-range interactions, are considered, obtaining a good agreement between simulated and experimental results. Our analysis shows that an efficient porting of regular homogeneous data-parallel applications on heterogeneous NOWs is possible. Particularly, the approach based on multiple processes per node turns out to be a straightforward and effective way for achieving very satisfying performance in almost all situations, even dealing with highly heterogeneous systems.

Performance analysis of SPMD algorithms on a network of workstations with virtual shared memory

We present a simple model to analyze and predict the performance of a given parallel application on a heterogeneous network of workstations with Virtual Shared Memory (VSM). Particularly, we deal with a switched Ethernet-based network and we use Network Linda as parallel framework; our analysis can however be applied to other virtual shared memory environments. Simulated and experimental results are in good agreement and the proposed method allows to evaluate how the efficiency varies with the task granularity and the degree of heterogeneity of the network.

SoRTSim: A High-Level Simulator for the Evaluation of QoS Models on Grid

We present SoRTSim, a high-level simulator designed to evaluate tools built over the middleware to support additional services besides the standard ones, particularly to allow the execution on Grid platforms of applications with strict QoS requirements, up to Soft Real-Time. SoRTSim has been designed to be quite general and to be easily customized for different tools and classes of applications.We describe the SoRTSim architecture and its implementation, highlighting its main features. The simulator allows the easy generation of a high number of abstract Grid scenarios, starting from a set of constraints; supports the definition of the various Grid entities and their behaviour; and makes simple the evaluation of performance, based on user-defined metrics and statistics.As a case study, we show how SoRTSim can be used to simulate SoRTGrid, a framework that we previously developed for managing time constraints on a Service-oriented Grid.