On using Tracer Driver for External Dynamic Process Observation

TL;DR

This paper introduces the concept of a 'full trace' and a tracer driver to enable simultaneous, flexible observation of dynamic processes through multiple analyzers, focusing on efficiency and practicality.

Contribution

It proposes the novel notion of a full trace and a tracer driver to facilitate multi-analyzer observation of processes, addressing efficiency and scalability issues.

Findings

Tracer driver restricts information flow for tractability.

Analysis of workload distribution between tracer and analyzers.

Application to constraint system resolution visualization.

Abstract

One is interested here in the observation of dynamic processes starting from the traces which they leave or those that one makes them produce. It is considered here that it should be possible to make several observations simultaneously, using a large variety of independently developed analyzers. For this purpose, we introduce the original notion of ``full trace'' to capture the idea that a process can be instrumented in such a way that it may broadcast all information which could ever be requested by any kind of observer. Each analyzer can then find in the full trace the data elements which it needs. This approach uses what has been called a "tracer driver" which completes the tracer and drives it to answer the requests of the analyzers. A tracer driver allows to restrict the flow of information and makes this approach tractable. On the other side, the potential size of a full trace seems to make the idea of full trace unrealistic. In this work we explore the consequences of this notion in term of potential efficiency, by analyzing the respective workloads between the (full) tracer and many different analyzers, all being likely run in true parallel environments. To illustrate this study, we use the example of the observation of the resolution of constraints systems (proof-tree, search-tree and propagation) using sophisticated visualization tools, as developed in the project OADymPPaC (2001-2004). The processes considered here are computer programs, but we believe the approach can be extended to many other kinds of processes.