[Back]

O. Scekic, S. Dustdar:
"The Social Compute Unit: Adaptability and Rewarding Mechanisms";
Talk: 5th Summer School on Service Oriented Computing 2011, Hersonissos, Crete, Greece; 2011-06-27 - 2011-07-03.

The Social Compute Unit (SCU) is a new concept that looks into the entire life-cycle of human expert teams in the context of Mixed Systems. The topic of our work is refining and extending the notion of SCU, with particular focus on the adaptability and rewarding mechanisms. The life-cycle of a SCU includes, among others, phases of team creation, execution, and dissolution.

During the execution phase the assignment is being executed by human contributors and monitored by the system. The running time of the execution phase can be specified in advance, or last arbitrarily long, until the assignment is finished or a satisfactory level of quality is reached. To support this kind of versatility, we propose dividing execution into iterative cycles. During an iteration the team composition remains fixed. Partial results and team reorganization are considered only at iteration's end. For simple tasks, a special case of single-iteration model can be adopted. We specify a number of benefits of such iterative approach.

Dissolution phase includes honoring of contractual obligations between the SCU-system and the contributors. This includes paying out of rewards or penalties. Their final amounts depend on the accumulated scores throughout the execution phase. We propose calculating and adjusting scores for each particular contributor at the end of every iteration. Scores depend on contributors' roles in the team, demonstrated effort and results obtained. Score calculation is subject to clear, rule-based policies specified as a domain-specific language. The purpose of incentives is to increase productivity while minimizing the reward money, by stimulating contributors to offer services for a smaller price. Penalties serve the purpose of maintaining the required QoS levels.