A contention aware EQS priority assignment heuristic for cohorts in DRTDBS

Abstract

The equal slack (EQS) priority assignment method has shown promising results with distributed real-time systems. As a result, the EQS is considered the first choice for distributed real-time database systems (DRTDBS). However, the integration of EQS with the DRTDBS comes at cost of many issues—intensive data contention due to dynamic cohort priority assignment, global deadlock, and wastage of resources due to cyclic restart. To address the above problems, this paper proposes a contention aware equal slack (CA-EQS) priority assignment heuristic. The CA-EQS heuristic reduces the data contention by utilizing the size of the cohort dependency. After the execution of the first cohort, the information about its size of dependency will be piggybacked with the message that initiates the execution of immediately next cohort of a parent transaction at some other site. The reason for going with the piggybacking approach is that the dependency size of the currently executing cohort not only depends on the data items locked by it but also on the data items that are locked by the cohorts that have already completed their execution before it. This way, a true data contention level can be assessed without any extra overhead (message and time). The CA-EQS solves the problem of deadlock and cyclic restart. Results from the simulation tool validate up to 9% drop in transactions miss ratio and up to 16% reduction in a number of transactions’ rollbacks by the CA-EQS heuristic over EQS, EQF, number of locks (NL), and most dependent transactions first (MDTF).