Partial mission aborting in work sharing systems

Mission aborting aims to survive a system and avert catastrophic damages caused by the loss of the system. However, the effectiveness of mission aborting depends heavily on the abort policy adopted. Thus, it is crucial to model the abort policy in the mission performance evaluation and further optimize the policy maximizing or minimizing the mission performance metric of interest. Rich research efforts have recently been expended in studying the abort policy, and all the existing models have assumed the full mission abort in the case of a predetermined condition being met. This paper makes novel contributions by modeling and optimizing a partial abort policy (PAP) for systems with multiple work sharing units. With the goal of balancing the amount of completed work and the risk of damages, the PAP proposed determines the number of units that should continue the primary mission (PM) whereas the rest of the operating units abort the PM and start the execution of a rescue procedure (RP). The traditional full abort policies appear as special cases of the proposed PAP model. A numerical algorithm based on probabilistic models is suggested for mission performance evaluation while considering dynamic operation conditions of units during PM and RP. Mission performance metrics evaluated include mission completion probability, damage avoidance probability, expected completed work fraction, and expected losses. The optimal PAP problem is further formulated and solved using the genetic algorithm with the objective of minimizing the expected losses. Influences of multiple model parameters on the mission performance metrics and optimization solutions are examined through an example of a multi-processor server system.

Automated summary

Mission aborting aims to prevent catastrophic damages by modeling and optimizing a partial abort policy (PAP) for systems with multiple work-sharing units. The proposed PAP balances completed work and risk of damages, contributing to mission performance evaluation and optimization.