Rainer Klar

Distributed performance monitoring: methods, tools, and applications

A method for analyzing the functional behavior and the performance of programs in distributed systems is presented. We use hybrid monitoring, a technique which combines advantages of both software monitoring and hardware monitoring. The paper contains a description of a hardware monitor and a software package (ZM4/SIMPLE) which make our concepts available to programmers, assisting them in debugging and tuning of their code. A short survey of related monitor systems highlights the distinguishing features of our implementation. As an application of our monitoring and evaluation system, the analysis of a parallel ray tracing program running on the SUPRENUM multiprocessor is described. It is shown that monitoring and modeling both rely on a common abstraction of a systems dynamic behavior and therefore can be integrated to one comprehensive methodology. This methodology is supported by a set of tools. >

Measuring simultaneous events in a multiprocessor system

In the hierarchically organized multiprocessor system EGPA, which has the structure of a pyramid, the performance of concurrent programs is studied. These studies are assisted by a hardware monitor (ZÄHLMONITOR III), which measures not only the activity and idle states of CPU and channels, but records the complete history of processes in the CPU and interleaved I/O activities. The applied method is distinguished from usual hardware measurements for two reasons: it puts together the a priori independent event-streams coming from the different processors to a well ordered single event stream and it records not only hardware but also software events. Most useful have been traces of software events, which give the programmer insight into the dynamic cooperation of distributed subtasks of his program. This paper describes the measurement method and its application to the analysis of the behaviour of a highly asynchronous parallel algorithm: the projection of contour lines from a given point of view and the elimination of hidden lines. This work is sponsored by the Bundesminister für Forschung und Technologie (German Federal Minister of Research and Technology).

ZÄHLMONITOR 4: Ein Monitorsystem für das Hardware- und Hybrid-Monitoring von Multiprozessor- und Multicomputer-Systemen

Die Beobachtung (das Monitoring) der inneren Ablaufe in Multiprozessor- und Multicomputer-Systemen ist ein hervorragendes Werkzeug, um die Ursachen fur die extern festgestellte Gesamtleistung solcher Systeme zu analysieren und damit die Voraussetzungen fur eine Leistungsverbesserung zu schaffen. Im vorliegenden Papier wird ein Monitorsystem vorgestellt, das speziell fur (auch raumlich verteilte) Multiprozessor- und Multicomputer-Systeme entworfen wurde. Seine wesentlichen Eigenschaften sind die Kombination von verteilter und zentraler Instrumentierung des Objektsystems, die systemweite Ermittlung von Ereignisreihenfolgen und die quellbezogene Auswertung von Ereignisspuren. Neben dem globalen Konzept werden vertiefende Untersuchungen dargestellt, und es wird von ersten Meserfahrungen berichtet.

VLSI-based monitoring of the inter-process-communication in multi-microcomputer systems with shared memory

Abstract To improve the performance of distributed systems, monitoring of communication activities is often the first step. In this paper a VLSI-chip named Chained Reference Address Comparator (CRAC) is presented that supports the monitoring of procedure- and message-oriented communication in distributed MMC systems with shared memory. This is done by monitoring the accesses of processors to the data structures allocated in shared memory that serve communication purposes. A key capability of the CRAC chip is the program controlled address comparison technique featuring automatic generation of reference addresses from address elements dynamically acquired by the CRAC chip from the data bus of the monitored system. This so-called chained address comparison makes it possible to monitor structure accesses to dynamic data structures, having an important function for the implementation of communication mechanisms. Due to its direct attachement to the processor address/data bus it is specifically designed for MMC systems with the Intel processor 8086.

Hardware-measurements of storage access conflicts in the processor array EGPA(1)

The contemporary general purpose computer of the Princeton type is limited in its performance by the speed of the available logic families. Many projects around the world try to achieve higher performance by using many processors connected via common memories. The hardware costs for the interconnection of many processors and memories has to be limited and a severe degradation resulting from memory access conflicts must be avoided. The EGPA-project uses the concept of restricted neighbourhoods to solve these problems. The processors and memories are arranged in a freely extensible cellular structure. The number of connections to processors and to memories are limited for all sizes of the array. In this paper we present measured results from the pilot system, which show whether memory conflicts will influence the speed of the individual processors. The results will be valid for larger arrays, too.

Instrumentierung eines Prozessor-feldes

Es wird dargestellt, wie ein Hardware-Monitor in einem Prozessorfeld mit begrenzten Nachbarschaften (Projekt EGPA) zur Messung und Bewertung eingesetzt wird. Im ersten Teil dieser Untersuchung wird beschrieben, welche Hilfsmittel erforderlich sind, wenn in allen Prozessoren simultane Realzeitmessungen erfolgen sollen, die den tatsachlichen Befehls- und Datenflus erfassen. Ein zweiter Teil zeigt, das erst besondere Eigenschaften der Prozessoren (AEG 80-60) die Erfassung von Software-Ereignissen mit Hardware-Messungen moglich macht.

Integrating Monitoring and Modeling to a Performance Evaluation Methodology

This paper presents a comprehensive methodology for monitoring and modeling parallel and distributed systems systematically. The integration of models, measurements, and evaluators to an efficient set of performance evaluation tools is described. Three typical tools are presented. One of them is the distributed hardware and hybrid monitor ZM4, another is the monitor independent and source related event trace interface POETITDL. Both were developed at the Universitat Erlangen-Nurnberg. As a modeling tool stochastic Petri-nets have been used. These tools have been used for analyzing the performance of multiprocessor and multicomputer systems. Here, they are applied in a case study for performance analysis and improvement of a communication subsystem prototype for B(roadband)-ISDN that was developed by IBM’s European Networking Center. The measurement results give some interesting hints concerning the prototype’s architecture which helped to improve the communication subsystem. Measuring the existing communication subsystem was accompanied by models for predicting the performance of modified ones.

Combining Gprof and Event—Driven Monitoring for Analyzing Distributed Programs：A Rough View of NCSA Mosaic

There are several purposes of analyzing a program: functional or performance analysis, debugging or, more recently, mapping a program to a new parallel or distributed architecture. In this paper, we introduce an effective method leading to the Execution Graph (EG) from a program. First, the Unix profiling tool Gprof is used to get the Execution Model (EM) of a C-program. Then the event-driven monitoring tool AICOS-SIMPLE is used to get the EG which includes not only the call graph but also the execution time table of the program. This method is suitable for analyzing modern distributed programs. As the example of the analysis, the well known HTTP protocol under the NCSA Mosaic is chosen. An EG of NCSA Mosaic on the routing level is given.

Status and Prospect of ZM4/SIMPLE/PEPP: an Event-oriented Evaluation Environment for Parallel and Distributed Programs

Debugging and performance evaluation of parallel and distributed programs can be facilitated by tools which consider a parallel program in terms of the dynamic flow of significant events and of their interaction.

Kausalität in Computersystemen

Um die Leistungsfahigkeit paralleler und verteilter Systeme wirklich nutzen zu konnen, mus man die komplizierten Interaktionen zwischen kooperierenden Prozessen verstehen. Dies erfordert unter anderem die Analyse von Kausalbeziehungen, welche ihrerseits den Rahmen fur die moglichen zeitlichen Reihenfolgen aller Ereignisse in einem parallelen und verteilten System bilden.