A Tracer for Debugging Multi-agent System Based on P-Q Signal Method

Abstract

We have developed RXF (Reflective Familiar) [1], a new programming environment for building a multi-agent system (MAS). We are interested in what types of functions are required when building MASs. An MAS in RXF consists of reflective agents, which are intelligent and autonomous agents based on reflection. RXF employs the constraint-logic programming for rapid prototyping, the multi-thread programming for concurrent programming, and the reflections for meta-level programming. On developing RXF we recognized the necessity of a new debugger for MASs because problems debugging concurrent processes makes debugging MASs more difficult. For example, a traditional tracer slows down only an agent debugged. As a result, a speed ratio of agents in an MAS becomes different from a speed ratio of agents without such a debugger. We proposed a multi-agent tracer that allows programmers to effectively debug MASs [2]. The tracer attempts to keep the speed ratio of agents in an MAS to solve the problem. The tracer is not enough robust against packet loss or network latency. We believe debugging MASs in unstable network environment. In this paper, we propose the P-Q signal method to improve the tracer in unstable network environment. The new method utilizes a P signal and a Q signal to inform non-debugged agents that the debugger suspends and resumes a debugged agent, respectively. The P and Q signals include suspend time and total suspend time to adjust the speed ratio. The new tracer can adjust a speed ratio of agents in an MAS more precisely by using P-Q signals.