mation enables Paradyn to dynamically select which per-mance problems at the lowest system levels, programmers formance data to collect (and when to collect it). The goal must examine low-level problems while maintaining refis for the tool to identify the parts of the program that are erences to the high-level source code. limiting performance instead of requiring the program-Paradyn allows high-level language programmers to mer to do it. view performance in terms of high-level Paradyn's Performance Consultant mod-objects (such as arrays and loops for dataule has a well-defined notion, called the W3 parallel Fortran) and automatically maps search model, of the types of performance P aradyn already the high-level information to low-level problems found in programs and of the varworks well in objects (such as nodes and messages). If ious components contained in the current several domains users want to view the low-level informaprogram. The Performance Consultant uses and measures pro-tion, Paradyn helps them relate perfor-W3 information to guide dynamic instrugrams running on mance data between levels. mentation placement and modification. heterogeneous combinations. Open interfaces for visualization Provide well-defined data abstractions Simple data abstractions can unify the design of a performance tool and simplify its organization. Paradyn uses two important abstractions-metric-focus grids and time histograms-in collecting, communicating, analyzing, and presenting performance data. A metric-focus grid is based on two lists (vectors) of information. The first vector is a list of performance metrics, such as CPU time, blocking time, message rates, I/O rates, or number of active processors. The second vector is a list of individual program components, such as a selection of procedures, processor nodes, disks, message channels, or barrier instances. The combination of these two vectors produces a matrix with each metric listed for each program component. The matrix elements can be single values, such as current rate, average, and sum, or time histograms, which record metric behavior as it varies over time. The time histogram is an important tool in recording time-varying data for long-running programs.

No abstract

Accurate performance diagnosis of parallel and distributed programs is a difficult and time-consuming task. We describe a new technique that uses historical performance data, gathered in previous executions of an application, to increase the effectiveness of automated performance diagnosis. We incorporate several different types of historical knowledge about the application's performance into an existing profiling tool, the Paradyn Parallel Performance Tool. We gather performance and structural data from previous executions of the same program, extract knowledge useful for diagnosis from this collection of data in the form of search directives, then input the directives to an enhanced version of Paradyn, which conducts a directed online diagnosis. Compared to existing approaches, incorporating historical data shortens the time required to identify bottlenecks, decreases the amount of unhelpful instrumentation, and improves the usefulness of the information obtained from a diagnostic session.

Event traces are required to correctly diagnose a number of performance problems that arise on today's highly parallel systems. Unfortunately, the collection of event traces can produce a large volume of data that is difficult, or even impossible, to store and analyze. One approach for compressing a trace is to identify repeating trace patterns and retain only one representative of each pattern. However, determining the similarity of sections of traces, i.e., identifying patterns, is not straightforward. In this paper, we investigate pattern-based methods for reducing traces that will be used for performance analysis. We evaluate the different methods against several criteria, including size reduction, introduced error, and retention of performance trends, using both benchmarks with carefully chosen performance behaviors, and a real application.

No abstract