Diamond quorum consensus for high capacity and efficiency in a replicated database system

Many quorum consensus protocols have been proposed for the management of re plicated data in a distributed environment. The advantages of a replicated database system over a non-replicated one include high availability and low response time. We note further that the multiple sites can act as multiple agents so that at any time, multiple requests can be handled in parallel. This feature leads to the desirable consequence of high workload capacity. In this paper, we define a new metric of read-capacity for this feature. We propose a new protocol called diamond quorum consensus which has two major properties that are superior to the previous protocols of majority, tree, grid, and hierarchical quorum consensus: (1) it has the highest read-capaci ty, (2) it has the smallest optimal read quorum size of 2. We show that the se two features are achievable without jeopardizing the availability. The s mall quorum size is a significant feature because it relates to the messagi ng cost. Few previous work on quorum consensus has discussed the handling o f partition failure, which in many cases will depend on the quorum consensu s protocol, we show how we can use the generalized virtual partition protoc ol to handle partition failure in the case of diamond quorum consensus.