SCI: NMI DEVELOPMENT: Performance Inside: Performance Monitoring and Diagnosis for NMI Software and Applications

University Of Chicago, Chicago IL

Investigators

Abstract

The proposal seeks to develop software and tools to automate management for the process of collecting, analyzing, and applying diverse performance information, and in so doing lead to the understanding of baseline performance behavior and failure conditions of distributed systems and applications. The tools will support the development of extended monitoring frameworks that enable the capture of performance data across the environment by using a set of sensors, or probes, widely deployed to gather basic data. The work will enable sensors to be managed and configured easily, so that the amount of data collected can be varied according to the situation. Sensors will be able to register into a common registry so that users can easily discover and select the sensors relevant to a specific application. Additional tools, such as archivers and prediction services, will be provided to store performance data and estimate future behaviors. Trigger services will be provided for early fault detection, and higher-level analysis tools for behavioral diagnosis. The project will produce (1) a NMI Sensor Toolkit, software that implements various simple sensor templates so that sensors can easily be incorporated into tools, applications, and other services to provide performance data to users; (2) performance-enabled NMI software, sensor-enhanced versions of various NMI components (e.g., Globus Toolkit's GRAM, GridFTP, RFT, MDS; MPICH-G2; OpenSSH; MyProxy); (3) sensor management tools, services and tools for archiving, analyzing, and controlling performance monitoring in a distributed environment; and (4) client tools (APIs, libraries, command line tools, utilities) that provide users and system administrators with easy and understandable access to performance data.