Vítor Duarte

Monitoring PVM Programs Using the DAMS Approach

Monitoring tools are fundamental components of a development environment as they provide basic support for performance evaluation, debugging, and program visualization. We describe our experiments with several monitoring tools for PVM, namely XPVM, developed at ORNL, Tape/PVM, developed at IMAG Lab, and DAMS, developed at UNL. These tools are compared and their use is described to support instrumentation and monitoring of a high level distributed language, PVM-Prolog, an extension to Prolog that provides an interface to PVM. This language is being used for the implementation of multi-agent systems, and it provides support for heterogeneous programs, built from C and Prolog components that communicate using PVM.

Using libPcap for monitoring distributed applications

The monitoring of parallel and distributed applications is a common approach for gathering information concerning program execution, for behavioral analysis of the application or of the supporting platform. The collected information can be used for the detection of errors, for performance evaluation and in its improvement. Several monitoring techniques have been explored by many different tools, depending on their goals and platforms used. In particular, the analysis made at the network level, also allows the study of the network and application protocols used by supporting middlewares and frameworks. In this paper, we report an experimental distributed monitoring tool that takes advantage of network traffic capture offered by the OS, in particular, using libPcap. Our objective was, first, to verify the possibility of capturing the traffic relating to a particular process and, based on this information, identify the other application processes and trigger the remote monitoring for those processes. This tool allows the transparent capture of the distributed applications interactions between its several processes. It is enough for the user to identify a single process and the remaining processes are discovered and monitored from the information gathered. More deep analysis can then be carried by any other tool supporting the pcap/tcpdump format, as for example, Wireshark.

Towards a Framework for Monitoring Grid Scientific Workflows

A typical workflow for a distributed application involves a large number of resources that can fail, including network, hardware and software components. Even when monitoring information from all these components is accessible, it is hard to determine how anomalies and failures during the application execution are related to a given workflow component. However the capability of receiving and interpreting intermediate results and interacting with applications plays a significant role for developing scientific experiments. Considering the complexity of implementation of distributed systems and the large scope of issues the monitoring system should cover, what analysis and planning is required to implement effective scientific grid workflow monitoring? We propose a multi-layer approach which focuses on a clear identification of the workflow-level monitoring abstractions. Through a clear separation between higher and lower level mechanisms, this approach will allow the specification of application monitoring requirements at workflow level, and their implementation upon distinct monitoring technologies, including the ones supported by existing grid middleware.

The DOTPAR Project: Towards a Framework Supporting Domain Oriented Tools for Parallel and Distributed Processing

We discuss the problem of building domain oriented environments by a composition of heterogeneous application components and tools. We describe several individual tools that support such environments, namely a distributed monitoring and control tool (DAMS), a process-based distributed debugger (PDBG) and a heterogeneous interconnection model (PHIS). We discuss our experience with the development of a Problem Oriented Environment in the domain of genetic algorithms, obtained by a composition of heterogeneous tools and application components.

Using a PC simulator to illustrate input-output programming techniques

We present our use of the Bochs PC emulator in a series of pratical assignments that, in a basic computer architecture course, introduce polling and interrupt-based input-output programming techniques.

A grid workload modeling approach for intelligent grid

As more and more grid applications are put into use, the performance information plays a more important role in evaluating the running status of the grid system. However, most existing monitoring tools only provide system-level performance data, which are often massive in storage and meanwhile contain either duplicate or missing, even erroneous raw data. In order to make the collected data more easily to understand and thus reusable, in this paper we discuss grid workload modeling approaches to better serve the grid environment. We extract three key workload objects from numerous monitoring entities as the backbone of the workload and put them into use to examine the completeness of a performance trace. We present a methodology for accessing and interpreting grid workload information. A grid workload, which is recorded as a structured collection of monitoring information that is kept in repository continuously updated as a reflection of the execution, is studied at three hierarchical levels. From a goal-oriented perspective, this systematic approach could help use the workload data to guide and improve the resource mappings and fine tune the application and system, opening the way towards an intelligent grid environment.