SYSTEM AND METHOD FOR EXECUTING, TRACKING AND RECOVERING LONG RUNNING COMPUTATIONS

摘要

A transaction description database represents long running computations as a set of computational steps with data flows therebetween. The description database defines each step's input and output signals, input condition criteria for creating an instance of the step, an application program associated with the step, and criteria for selecting a resource to execute the step. A flow controller controls the process of executing instances of each defined type of long running transaction. Execution of a long running transaction begins when a corresponding set of externally generated input event signals are received by the flow controller. During execution of a long running transaction, each step of the transaction is instantiated only when a sufficient set of input signals is received to execute that step. At that point an instance of the required type of step is created and then executed by a selected resource. After termination of a step, output signals from the step are converted into input event signals for other steps in the long running transaction in accordance with data stored in the transaction description database. Each step executes an application program and is treated as an individual computation insofar as durable storage of its computational results. Log records are durably stored upon instantiation, execution and termination of each step of a long running transaction, and output event signals are also logged, thereby durably storing sufficient data to recover a long running transaction with virtually no loss of the work that was accomplished prior to a system failure. $$