Transactional memory (TM) simplifies parallel programming by guaranteeing that transactions appear to execute atomically and in isolation. Implementing these properties includes providing data version management for the simultaneous storage of both new (visible if the transaction commits) and old (retained if the transaction aborts) values. Most (hardware) TM systems leave old values “in place” (the target memory address) and buffer new values elsewhere until commit. This makes aborts fast, but penalizes (the much more frequent) commits. In this paper, we present a new implementation of transactional memory, Log-based Transactional Memory (LogTM), that makes commits fast by storing old values to a per-thread log in cacheable virtual memory and storing new values in place. LogTM makes two additional contributions. First, LogTM extends a MOESI directory protocol to enable both fast conflict detection on evicted blocks and fast commit (using lazy cleanup). Second, LogTM handles aborts in (library) software with little performance penalty. Evaluations running micro- and SPLASH-2 benchmarks on a 32way multiprocessor support our decision to optimize for commit by showing that only 1-2 % of transactions abort. 1.

Copyright © 2006, by the author(s).

Transactional memory (TM) promises to substantially reduce the difficulty of writing correct, efficient, and scalable concurrent programs. But “bounded ” and “best-effort” hardware TM proposals impose unreasonable constraints on programmers, while more flexible software TM implementations are considered too slow. Proposals for supporting “unbounded” transactions in hardware entail significantly higher complexity and risk than best-effort designs. We introduce Hybrid Transactional Memory (HyTM), an approach to implementing TM in software so that it can use best-effort hardware TM (HTM) to boost performance but does not depend on HTM. Thus programmers can develop and test transactional programs in existing systems today, and can enjoy the performance benefits of HTM support when it becomes available. We describe our prototype HyTM system, comprising a compiler and a library. The compiler allows a transaction to be attempted using best-effort HTM, and retried using the software library if it fails. We have used our prototype to “transactify ” part of the Berkeley DB system, as well as several benchmarks. By disabling the optional use of HTM, we can run all of these tests on existing systems. Furthermore, by using a simulated multiprocessor with HTM support, we demonstrate the viability of the HyTM approach: it can provide performance and scalability approaching that of an unbounded HTM implementation, without the need to support all transactions with complicated HTM support.

Transactional memory (TM) is an appealing abstraction for programming multi-core systems. Potential target applications for TM, such as business software and video games, are likely to involve complex data structures and large transactions, requiring specific software solutions (STM). So far, however, STMs have been mainly evaluated and optimized for smaller scale benchmarks. We revisit the main STM design choices from the perspective of complex workloads and propose a new STM, which we call SwissTM. In short, SwissTM is lock- and word-based and uses (1) optimistic (commit-time) conflict detection for read/write conflicts and pessimistic (encounter-time) conflict detection for write/write conflicts, as well as (2) a new two-phase contention manager that ensures the progress of long transactions while inducing no overhead on short ones. SwissTM outperforms state-of-theart STM implementations, namely RSTM, TL2, and TinySTM, in our experiments on STMBench7, STAMP, Lee-TM and red-black tree benchmarks. Beyond SwissTM, we present the most complete evaluation to date of the individual impact of various STM design choices on the ability to support the mixed workloads of large applications.

We describe DSTM2, a Java TM software library that provides a flexible framework for implementing object-based software transactional memory (STM). The library uses transactional factories to transform sequential (unsynchronized) classes into atomic (transactionally synchronized) ones, providing a substantial improvement over the awkward programming interface of our previous DSTM library. Furthermore, researchers can experiment with alternative STM mechanisms by providing their own factories. We demonstrate this flexibility by presenting two factories: one that uses essentially the same mechanisms as the original DSTM (with some enhancements), and another that uses a completely different approach. Because DSTM2 is packaged as a Java library, a wide range of programmers can easily try it out, and the community can begin to gain experience with transactional programming. Furthermore, researchers will be able to use the body of transactional programs that arises from this community experience to test and evaluate different STM mechanisms simply by supplying new transactional factories. We believe that this flexible approach will help to build consensus about the best ways to implement transactions, and will avoid the premature “lock-in ” that may arise if STM mechanisms are baked into compilers before such experimentation is done.

Concurrency-related bugs may happen when multiple threads access shared data and interleave in ways that do not correspond to any sequential execution. Their absence is not guaranteed by the traditional notion of “data race ” freedom. We present a new definition of data races in terms of 11 problematic interleaving scenarios, and prove that it is complete by showing that any execution not exhibiting these scenarios is serializable for a chosen set of locations. Our definition subsumes the traditional definition of a data race as well as high-level data races such as stale-value errors and inconsistent views. We also propose a language feature called atomic sets of locations, which lets programmers specify the existence of consistency properties between fields in objects, without specifying the properties themselves. We use static analysis to automatically infer those points in the code where synchronization is needed to avoid data races under our new definition. An important benefit of this approach is that, in general, far fewer annotations are required than is the case with existing approaches such as synchronized blocks or atomic sections. Our implementation successfully inferred the appropriate synchronization for a significant subset of Java’s Standard Collections framework.

In recent software transactional memory proposals, a contention manager module is responsible for ensuring that the system as a whole makes progress. A number of contention manager algorithms have been proposed and empirically evaluated. In this paper we lay some foundations for a theory of contention management. We present the greedy contention manager, the first to combine non-trivial provable properties with good practical performance. In a model where transaction delays are finite, the greedy manager guarantees that every transaction commits within a bounded time, and the time to complete n concurrent transactions that share s objects is within a factor of s(s + 1) + 2 of the time that would have been taken by an optimal offline list scheduler. No contention manager reviewed in the literature satisfies both the properties. Benchmark results convey our claim of the practicality of the greedy manager.

Mutual exclusion locks remain the de facto mechanism for concurrency control on shared-memory data structures. However, their apparent simplicity is deceptive: it is hard to design scalable locking strategies because locks can harbour problems such as priority inversion, deadlock and convoying. Furthermore, scalable lock-based systems are not readily composable when building compound operations. In looking for solutions to these problems, interest has developed in nonblocking systems which have promised scalability and robustness by eschewing mutual exclusion while still ensuring safety. However, existing techniques for building non-blocking systems are rarely suitable for practical use, imposing substantial storage overheads, serialising non-conflicting operations, or requiring instructions not readily available on today’s CPUs. In this paper we present three APIs which make it easier to develop non-blocking implementations of arbitrary data structures. The first API is a multi-word compare-and-swap operation (MCAS) which atomically updates a set of memory locations. This can be used to advance a data structure from one consistent state to another. The second API is a word-based software transactional memory (WSTM) which can allow sequential code to be re-used more directly than with MCAS and which provides better scalability when locations are being read rather than being

Abstract — Transactional memory has great potential for simplifying multithreaded programming by allowing programmers to specify regions of the program that must appear to execute atomically. Transactional memory implementations then optimistically execute these transactions concurrently to obtain high performance. This work shows that the same atomic guarantees that give transactions their power also have unexpected and potentially serious negative effects on programs that were written assuming narrower scopes of atomicity. We make four contributions: (1) we show that a direct translation of lock-based critical sections into transactions can introduce deadlock into otherwise correct programs, (2) we introduce the terms strong atomicity and weak atomicity to describe the interaction of transactional and non-transactional code, (3) we show that code that is correct under weak atomicity can deadlock under strong atomicity, and (4) we demonstrate that sequentially composing transactional code can also introduce deadlocks. These observations invalidate the intuition that transactions are strictly safer than lock-based critical sections, that strong atomicity is strictly safer than weak atomicity, and that transactions are always composable. I.

High performance parallel programs are currently difficult to write and debug. One major source of difficulty is protecting concurrent accesses to shared data with an appropriate synchronization mechanism. Locks are the most common mechanism but they have a number of disadvantages, including possibly unnecessary serialization, and possible deadlock. Transactional memory is an alternative mechanism that makes parallel programming easier. With transactional memory, a transaction provides atomic and serializable operations on an arbitrary set of memory locations. When a transaction commits, all operations within the transaction become visible to other threads. When it aborts, all operations in the transaction are rolled back. Transactional