Trusted Recovery in Distributed Systems

Abstract

This chapter focuses on the techniques to assess and repair the damage caused in a distributed database. Distributed databases are widely used in a variety of real world data intensive applications, and we need efficient damage assessment and repair algorithms to make them resilient to attacks. In Chapters 4 and 5, we have developed a suite of trusted recovery algorithms for centralized database systems, however, the complexity of distributed database systems caused by data partition, distributed transaction processing, and failures suggests that centralized trusted recovery algorithms cannot be (directly) applied in distributed database systems and new distributed damage assessment and repair algorithms are needed. First, since data are partitioned and stored on multiple sites, damage assessment and repair must be performed at multiple sites. Second, since one distributed transaction may access data stored at multiple sites, damage assessment, which is now based on the affecting relationships among distributed transactions, needs to coordinate and synthesize the affecting relationships among the subtransactions at each site. Third, distributed damage assessment and repair need to tolerate not only site failures but also communication failures.