A Blockchain-Based Decentralized Framework for Fair Data Processing

The blockchain has been considered as a new decentralized computing paradigm that has great potential to meet various computing needs. Considering a private network (such as data center) where incentive mechanisms are not required, this paper innovatively remolds the transaction-recording blockchain for decentralized data processing. In our design, workers have different processing capacity and tasks have different resource requirements. Workers first get task information from the blockchain and then process tasks locally, and next perform the proof of useful work (PoUW) consensus to compete for a scheduler, according to the number of the consumed CPU instructions in data processing. The scheduler is responsible for dispatching task information to the blockchain. A salient feature of our decentralized data processing is that workers actively select tasks, instead of passively receiving tasks as in a centralized framework. This will lead to collisions (i.e., multiple workers select the same task). To alleviate the collisions and provide the max-min fairness of data-processing, we propose a modified fair queue (called M-FQ) algorithm for the scheduler, as well as a fair task selection with collision avoidance (called Fair-CA) scheme for workers. Extensive simulations verify that our framework can well balance the fairness and the collision, while achieving as high throughput and good fairness as centralized frameworks. This study is the first attempt toward designing a general decentralized computing framework.