Optimizing dynamic memory allocation for software transactional memory systems

Abstract

The widespread use of multicore processors has renewed the interest in parallelism. In particular, a new concurrent programming paradigm known as transactional memory has attracted the attention of both Academy and Industry. The major difference from the transactional approach to the classicalsynchronization methods such as locks is on its higher abstraction level, an essential element for the development and maintenance of concurrent applications. However, software implementation of the transactional framework are largely criticized for its low performance. As a result, much of the research conducted in the last years has focused on the improvement of transactionalalgorithms so that the performance overhead can be reduced. Following this research line, this project aims at improving the efficiency of the software implementations of the transactional mechanism by exploiting an aspect that received little attention: dynamic memory allocation performed among transactions. Initially we intend to analyze the performance overhead due to the dynamic allocator and, based on this analysis, we seek to propose an allocation methodology especially devised to the transactional memory domain. (AU)