The butterfly barrier

Brooks, Eugene D.

doi:10.1007/bf01407877

Cited by 96 publications

(50 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Most use some form of tree to gather and scatter information [6,10,14,16]; the butterfly and dissemination barriers of Brooks [2] and of Hensgen, Finkel, and Manber [6] use a symmetric pattern of synchronization operations that resembles an FFT or parallel prefix computation. The butterfly and dissemination barriers perform a total of O (P log P) writes to shared locations, but only O (log P) on their critical paths.…”

Section: Unnecessary Waitingmentioning

confidence: 99%

“…In the absence of contention, a remote memory reference (read) takes about 4 µs, roughly 5 times as long as a local reference. reinitialize (previous_instance) previous_instance := current_instance current_instance := if current_instance = &instances [2] then &instances[0] else current_instance + 1 Figure 6: A fuzzy adaptive combining tree barrier with local-only spinning and breadth-first process wakeup.…”

Section: Experimental Environmentmentioning

confidence: 99%

See 1 more Smart Citation

Fast, contention-free combining tree barriers for shared-memory multiprocessors

Scott

Mellor-Crummey

1994

Int J Parallel Prog

View full text Add to dashboard Cite

Counter-based algorithms for busy-wait barrier synchronization execute in time linear in the number of synchronizing processes. This time can be made logarithmic in the number of processes by adopting algorithms based on trees or FFT-like synchronization patterns. As an additional improvement, Gupta and Hill [5] have proposed an adaptive combining tree barrier that exploits non-uniformity in inter-barrier computation times: processes begin to leave the barrier in time logarithmic in the number of processes when all processes arrive at once, but in constant time after the arrival of the last process when arrival times are skewed. Building on earlier work [4], Gupta and Hill present both regular and fuzzy versions of their barrier. The fuzzy version allows a process to perform useful work between the point at which it notifies other processes of its arrival at the barrier and the point at which it waits for all other processes to arrive.Unfortunately, like many forms of busy-wait synchronization, adaptive combining tree barriers as originally devised can induce large amounts of memory and interconnect contention in shared-memory multiprocessors, seriously degrading performance. They also perform a comparatively large amount of work at every tree node, raising the possibility that the constant factors in their execution time may be unacceptably high on machines of reasonable size. To address these problems, we present a new adaptive combining tree barrier, with fuzzy variant, that achieves significant speed improvements by spinning only on locally-accessible locations, and by using atomic fetch_and_store operations to avoid explicit locking of tree nodes. We also present a version of this barrier (again with fuzzy variant) that employs breadth-first wakeup of processes to reduce context switching when processors are multiprogrammed. We compare the performance of these new algorithms to that of other fast barriers on a 64-node BBN Butterfly 1 multiprocessor and on a 35-node BBN TC2000. Results suggest that adaptation is of little benefit, but that the combination of fuzziness with tree-style synchronization is of significant practical importance: fuzzy combining tree barriers with local-only spinning outperform all known alternatives on the TC2000 when the amount of fuzzy computation exceeds about 10% of the time between barriers.

show abstract

Section: Unnecessary Waitingmentioning

confidence: 99%

Section: Experimental Environmentmentioning

confidence: 99%

Fast, contention-free combining tree barriers for shared-memory multiprocessors

Scott

Mellor-Crummey

1994

Int J Parallel Prog

View full text Add to dashboard Cite

show abstract

“…There are four main classes of algorithms: masterslave [7], all-to-all [8], tree-based [7,9], butterfly [10]. Among them, the all-to-all algorithm takes a distributed solution.…”

Section: Introduction and Related Workmentioning

confidence: 99%

Cooperative communication for efficient and scalable all-to-all barrier synchronization on mesh-based many-core NoCs

Chen

Jantsch

et al. 2014

IEICE Electron. Express

View full text Add to dashboard Cite

On many-core Network-on-Chips (NoCs), communication is on the critical path of system performance and contended synchronization requests may cause large performance penalty. Different from conventional algorithm-based approaches, the paper addresses the barrier synchronization problem from the angle of optimizing its communication performance and proposes cooperative communication as a means to achieve efficient and scalable all-to-all barrier synchronization on mesh-based many-core NoCs. With the cooperative communication, routers collaborate with one another to accomplish a fast barrier synchronization task. The cooperative communication is implemented in our router at low cost. Through comparative experiments, our approach evidently exhibits high efficiency and good scalability.

show abstract

“…There are four main classes of algorithms: master-slave [3], all-to-all [4], tree-based [3,5], and butterfly [6]. Recently, as a single chip is being able to integrate many cores, barrier synchronization becomes a critical concern in single-chip systems due to its impact on application performance.…”

Section: Introduction and Related Workmentioning

confidence: 99%

Cooperative communication based barrier synchronization in on-chip mesh architectures

Chen

Jantsch

et al. 2011

IEICE Electron. Express

View full text Add to dashboard Cite

Abstract:We propose cooperative communication as a means to enable efficient and scalable barrier synchronization on mesh-based manycore architectures. Our approach is different from but orthogonal to conventional algorithm-based optimizations. It relies on collaborating routers to provide efficient gather and multicast communication. In conjunction with a master-slave algorithm, it exploits the mesh regularity to achieve efficiency. The gather and multicast functions have been implemented in our router. Synthesis results suggest marginal area overhead. With synthetic and benchmark experiments, we show that our approach significantly reduces synchronization completion time and increases speedup.

show abstract

The butterfly barrier

Cited by 96 publications

References 6 publications

Fast, contention-free combining tree barriers for shared-memory multiprocessors

Fast, contention-free combining tree barriers for shared-memory multiprocessors

Cooperative communication for efficient and scalable all-to-all barrier synchronization on mesh-based many-core NoCs

Cooperative communication based barrier synchronization in on-chip mesh architectures

Contact Info

Product

Resources

About