Exploiting non-determinism for reliability of mobile agent systems

Fault-tolerant mobile agent execution

“…In other words, when the agent execution is done, either a number of messages or another agent need to be sent to commit and/or abort all operations. This additional overhead, that, to our understanding, seems not to have been taken into account in the performance evaluation in [18], [26], results in reduced overall system throughput. In contrast, commit-after-stage approaches do not need to run a global commit/abort protocol.…”

“…In contrast, commit-at-dest approaches generally commit the modifications only at the end of the entire agent execution. Our approach, as well as the approaches in [3], [15], [20], [23], is commit-after-stage approaches, whereas [18], [26] are part of commit-at-dest approaches.…”

“…Incorrect failure suspicion may lead to duplicate agents and thus to a violation of the exactly-once execution property. Generally, duplicate agents are detected only at the agent destination [18], [26]. Consequently, accessed data items have to remain locked until the execution of the current agent finishes.…”

“…In the place-dependent approach, the FTE is provided by the mobile agent platform, e.g., [18], [26]. Here, fault tolerance is built into the places and a new instance of the FTE is created and executes at every stage of the agent execution.…”

“…Although fault-tolerant mobile agent approaches exist, we have found that current solutions are either complex and thus difficult to prove correct [23], [3], make limiting assumptions such as correct failure detection [15] or strict timing constraints [20], or only release resources at the end of the mobile agent execution [18], [26]. In this article, we model fault-tolerant mobile agent execution as a sequence of agreement problems.…”