Efficient Support for In-Place Metadata in Transactional Memory

Abstract

Abstract. Implementations of Software Transactional Memory (STM) algorithms associate metadata with the memory locations accessed during a transaction's lifetime. This metadata may be stored either in-place, by wrapping every memory cell in a container that includes the memory cell itself and the corresponding metadata; or out-place (also called external), by resorting to a mapping function that associates the memory cell address with an external table entry containing the corresponding metadata. The implementation techniques for these two approaches are very different and each STM framework is usually biased towards one of them, only allowing the efficient implementation of STM algorithms following that approach, hence inhibiting the fair comparison with STM algorithms falling into the other. In this paper we introduce a technique to implement in-place metadata that does not wrap memory cells, thus overcoming the bias by allowing STM algorithms to directly access the transactional metadata. The proposed technique is available as an extension to the DeuceSTM framework, and enables the efficient implementation of a wide range of STM algorithms and their fair (unbiased) comparison in a common STM infrastructure. We illustrate the benefits of our approach by analyzing its impact in two popular TM algorithms with two different transactional workloads, TL2 and multi-versioning, with bias to out-place and in-place respectively.

Efficient support for in‐place metadata in Java software transactional memory

“…A preliminary version of the work described in this paper was published in the Euro‐Par 2012 conference .…”

“…This paper reports on an extension of our previous work , which proposes a novel approach to support the in‐place metadata strategy without making use of the decorator pattern, and thoroughly evaluates its implementation in Deuce. This extension allows the efficient implementation of algorithms requiring a one‐to‐one relation between metadata and memory locations, such as multiversion algorithms.…”

“…To implement an algorithm in Deuce that requires a one-to-one relation between metadata and memory locations, such as a multiversion algorithm, one needs to use an external table that handles collisions, which significantly degrades the throughput of the algorithm. This paper reports on an extension of our previous work [7], which proposes a novel approach to support the in-place metadata strategy without making use of the decorator pattern, and thoroughly evaluates its implementation in Deuce. This extension allows the efficient implementation of algorithms requiring a one-to-one relation between metadata and memory locations, such as multiversion algorithms.…”