Real Time Concurrency Control for Data Intensive Applications

Abstract

A new class of data-intensive applications has received wide attention. In these applications, monitoring junctions can be performed by executing pre-registered transactions periodically using data from external sources in the form of continuous data streams and producing updates as temporal data in the database. For this type of continuous transactions, the data access pattern has to be declared before submission. To keep the database fresh, these transactions must be processed in a timely fashion. However, it is also desirable to execute other one-time transactions that may use these temporal data within their deadlines. Traditional DBMS are not designed to support such applications. The theme of this paper is to propose new concurrency control protocols to process these types of transactions with competing requirements. The challenge of this work is to balance database freshness and transaction performance. Although there is a number of similar previous works in the literature, they did not consider the presence of continuous transactions and therefore the issue of data dependency between these two types of transactions is not addressed. The simulation results show that our protocols can significantly improve the performance of one-time transactions while keeping the database fresh.