Automatic performance visualization of distributed real-time systems

For distributed real-time systems, adequate profiling tools are exceedingly rare. The sheer variety and low-level nature of these systems impede the adoption of standard, general-purpose tools for performance analysis and visualization. Although much research has been devoted to profiling parallel clusters and supercomputers, the literature virtually ignores the real-time domain. Correspondingly, a handful of commercial tools is available for profiling realtime software, but they invariably make a single-node assumption and are unable to cope with distributed environments. We examine the state of performance analysis and discuss why profilers are conspicuously absent in the field of distributed real-time systems. We then explore how developers of these systems could benefit from graphical profiling tools with automatic instrumentation and data collection. Toward that end, we demonstrate the prototype of a performance visualization tool called "Bacara", the second addition to our suite of tools for visual analysis of distributed real-time systems, or VADR (vamacr'der)

1. Background

Over the last twenty years, software profilers have become commonplace on the programmer's workbench. These tools help focus optimization efforts on selecting the right algorithms and tuning time-critical code, often resulting in large overall performance gains. Profiling software can also help reveal performance problems that may lie far from where programmer's intuition expects them to be, saving development time that would be wasted chasing down false bottlenecks.