arXiv subjects

Trevor Brown

Publications and source records attributed to Trevor Brown.

27 records · Page 2Linked to original sources

arXiv2018

Relaxed Schedulers Can Efficiently Parallelize Iterative Algorithms

There has been significant progress in understanding the parallelism inherent to iterative sequential algorithms: for many classic algorithms, the depth of the dependence structure is now well understood, and scheduling techniques have been developed to exploit this shallow dependence structure for efficient parallel implementations. A related, applied research strand has studied methods by which certain iterative task-based algorithms can be efficiently parallelized via relaxed concurrent priority schedulers. These allow for high concurrency when inserting and removing tasks, at the cost of executing superfluous work due to the relaxed semantics of the scheduler. In this work, we take a step towards unifying these two research directions, by showing that there exists a family of relaxed priority schedulers that can efficiently and deterministically execute classic iterative algorithms such as greedy maximal independent set (MIS) and matching. Our primary result shows that, given a randomized scheduler with an expected relaxation factor of $k$ in terms of the maximum allowed priority inversions on a task, and any graph on $n$ vertices, the scheduler is able to execute greedy MIS with only an additive factor of poly($k$) expected additional iterations compared to an exact (but not scalable) scheduler. This counter-intuitive result demonstrates that the overhead of relaxation when computing MIS is not dependent on the input size or structure of the input graph. Experimental results show that this overhead can be clearly offset by the gain in performance due to the highly scalable scheduler. In sum, we present an efficient method to deterministically parallelize iterative sequential algorithms, with provable runtime guarantees in terms of the number of executed tasks to completion.

cs.DS↗

arXiv2017

A General Technique for Non-blocking Trees

We describe a general technique for obtaining provably correct, non-blocking implementations of a large class of tree data structures where pointers are directed from parents to children. Updates are permitted to modify any contiguous portion of the tree atomically. Our non-blocking algorithms make use of the LLX, SCX and VLX primitives, which are multi-word generalizations of the standard LL, SC and VL primitives and have been implemented from single-word CAS. To illustrate our technique, we describe how it can be used in a fairly straightforward way to obtain a non-blocking implementation of a chromatic tree, which is a relaxed variant of a red-black tree. The height of the tree at any time is $O(c+ \log n)$, where $n$ is the number of keys and $c$ is the number of updates in progress. We provide an experimental performance analysis which demonstrates that our Java implementation of a chromatic tree rivals, and often significantly outperforms, other leading concurrent dictionaries.

cs.DC↗

arXiv2017

Pragmatic Primitives for Non-blocking Data Structures

We define a new set of primitive operations that greatly simplify the implementation of non-blocking data structures in asynchronous shared-memory systems. The new operations operate on a set of Data-records, each of which contains multiple fields. The operations are generalizations of the well-known load-link (LL) and store-conditional (SC) operations called LLX and SCX. The LLX operation takes a snapshot of one Data-record. An SCX operation by a process $p$ succeeds only if no Data-record in a specified set has been changed since $p$ last performed an LLX on it. If successful, the SCX atomically updates one specific field of a Data-record in the set and prevents any future changes to some specified subset of those Data-records. We provide a provably correct implementation of these new primitives from single-word compare-and-swap. As a simple example, we show how to implement a non-blocking multiset data structure in a straightforward way using LLX and SCX.

cs.DC↗

arXiv2017

B-slack trees: Highly Space Efficient B-trees

B-slack trees, a subclass of B-trees that have substantially better worst-case space complexity, are introduced. They store $n$ keys in height $O(\log_b n)$, where $b$ is the maximum node degree. Updates can be performed in $O(\log_{\frac b 2} n)$ amortized time. A relaxed balance version, which is well suited for concurrent implementation, is also presented.

cs.DS↗

arXiv2017

Range Queries in Non-blocking $k$-ary Search Trees

We present a linearizable, non-blocking $k$-ary search tree ($k$-ST) that supports fast searches and range queries. Our algorithm uses single-word compare-and-swap (CAS) operations, and tolerates any number of crash failures. Performance experiments show that, for workloads containing small range queries, our $k$-ST significantly outperforms other algorithms which support these operations, and rivals the performance of a leading concurrent skip-list, which provides range queries that cannot always be linearized.

cs.DC↗

arXiv2017

Techniques for Constructing Efficient Lock-free Data Structures

Building a library of concurrent data structures is an essential way to simplify the difficult task of developing concurrent software. Lock-free data structures, in which processes can help one another to complete operations, offer the following progress guarantee: If processes take infinitely many steps, then infinitely many operations are performed. Handcrafted lock-free data structures can be very efficient, but are notoriously difficult to implement. We introduce numerous tools that support the development of efficient lock-free data structures, and especially trees.

cs.DC↗

arXiv2017

Reclaiming memory for lock-free data structures: there has to be a better way

Memory reclamation for lock-based data structures is typically easy. However, it is a significant challenge for lock-free data structures. Automatic techniques such as garbage collection are inefficient or use locks, and non-automatic techniques either have high overhead, or do not work for many data structures. For example, subtle problems can arise when hazard pointers, one of the most common non-automatic techniques, are applied to many lock-free data structures. Epoch based reclamation (EBR), which is by far the most efficient non-automatic technique, allows the number of unreclaimed objects to grow without bound, because one crashed process can prevent all other processes from reclaiming memory. We develop a more efficient, distributed variant of EBR that solves this problem. It is based on signaling, which is provided by many operating systems, such as Linux and UNIX. Our new scheme takes $O(1)$ amortized steps per high-level operation on the data structure and $O(1)$ steps in the worst case each time an object is removed from the data structure. At any point, $O(mn^2)$ objects are waiting to be freed, where $n$ is the number of processes and $m$ is a small constant for most data structures. Experiments show that our scheme has very low overhead: on average 10\%, and at worst 28\%, for a balanced binary search tree over many thread counts, operation mixes and contention levels. Our scheme also outperforms a highly tuned implementation of hazard pointers by an average of 75\%. Typically, memory reclamation is tightly woven into lock-free data structure code. To improve modularity and facilitate the comparison of different memory reclamation schemes, we also introduce a highly flexible abstraction. It allows a programmer to easily interchange schemes for reclamation, object pooling, allocation and deallocation with virtually no overhead, by changing a single line of code.

cs.DC↗

arXiv2017

A Template for Implementing Fast Lock-free Trees Using HTM

Algorithms that use hardware transactional memory (HTM) must provide a software-only fallback path to guarantee progress. The design of the fallback path can have a profound impact on performance. If the fallback path is allowed to run concurrently with hardware transactions, then hardware transactions must be instrumented, adding significant overhead. Otherwise, hardware transactions must wait for any processes on the fallback path, causing concurrency bottlenecks, or move to the fallback path. We introduce an approach that combines the best of both worlds. The key idea is to use three execution paths: an HTM fast path, an HTM middle path, and a software fallback path, such that the middle path can run concurrently with each of the other two. The fast path and fallback path do not run concurrently, so the fast path incurs no instrumentation overhead. Furthermore, fast path transactions can move to the middle path instead of waiting or moving to the software path. We demonstrate our approach by producing an accelerated version of the tree update template of Brown et al., which can be used to implement fast lock-free data structures based on down-trees. We used the accelerated template to implement two lock-free trees: a binary search tree (BST), and an (a,b)-tree (a generalization of a B-tree). Experiments show that, with 72 concurrent processes, our accelerated (a,b)-tree performs between 4.0x and 4.2x as many operations per second as an implementation obtained using the original tree update template.

cs.DC↗

arXiv2017

Reuse, don't Recycle: Transforming Lock-free Algorithms that Throw Away Descriptors

In many lock-free algorithms, threads help one another, and each operation creates a descriptor that describes how other threads should help it. Allocating and reclaiming descriptors introduces significant space and time overhead. We introduce the first descriptor abstract data type (ADT), which captures the usage of descriptors by lock-free algorithms. We then develop a weak descriptor ADT which has weaker semantics, but can be implemented significantly more efficiently. We show how a large class of lock-free algorithms can be transformed to use weak descriptors, and demonstrate our technique by transforming several algorithms, including the leading k-compare-and-swap (k-CAS) algorithm. The original k-CAS algorithm allocates at least k+1 new descriptors per k-CAS. In contrast, our implementation allocates two descriptors per process, and each process simply reuses its two descriptors. Experiments on a variety of workloads show significant performance improvements over implementations that reclaim descriptors, and reductions of up to three orders of magnitude in peak memory usage.

cs.DC↗