• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Parallel and Distributed Computing and Networks >Speculative Prefetching of Optional Locks in Distributed Systems
【24h】

Speculative Prefetching of Optional Locks in Distributed Systems

机译：分布式系统中可选锁的推测性预取

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

We present a family of methods for speeding up distributed locks by exploiting the uneven distribution of both temporal and spatial locality of access behaviour of many applications. In the worst case, some of our methods will not produce higher network latencies than equivalent conventional distributed locking methods. In best case, the total number of messages can be constantly bounded, approximating the impression that no network latencies exist at all. Measurements and simulations based on variants of TPC database benchmarks show that hit rates enabled by speculative prefetching of optional locks appear to be similar to hit rates of conventional data caches, typically in the range from 90% to 99%. Thus overall speedup factors of 10 or more for the average latencies of distributed locks are possible. Compared to purely temporal prefetching, adding exploitation of spatial locality may significantly improve performance, typically by factors of 2 or more. We discuss some implications for the construction of distributed systems. For the class of programs well-suited for distributed systems, network latencies will nearly vanish, blurring performance differences between local and distributed systems. For program classes exposing poor locality, there is likely no help independent from distributed computing paradigms. We explain how the communica-tion paradigm can be efficiently implemented on top of distributed shared memory (DSM) using region locks. Thus we believe that the DSM paradigm will become more attractive than explicit communication (e.g. RPC, CORBA) for the construction of distributed applications.
机译：我们提出了一系列利用许多应用程序的访问行为的时间和空间局部性的不均匀分布来加速分布式锁的方法。在最坏的情况下，我们的某些方法不会产生比等效的常规分布式锁定方法更高的网络延迟。在最好的情况下，消息的总数可以不断限制，近似于根本没有网络等待时间的印象。根据TPC数据库基准的变体进行的测量和模拟显示，通过推测性预取可选锁实现的命中率似乎与常规数据缓存的命中率相似，通常在90％到99％的范围内。因此，对于分布式锁的平均等待时间，总加速因子为10或更高是可能的。与纯时间预取相比，增加对空间局部性的利用可以显着提高性能，通常提高2倍或更多。我们讨论了一些对分布式系统构建的启示。对于非常适合分布式系统的一类程序，网络等待时间几乎消失，从而模糊了本地系统和分布式系统之间的性能差异。对于暴露局部性差的程序类，可能没有独立于分布式计算范例的帮助。我们将说明如何使用区域锁在分布式共享内存（DSM）之上有效地实现通信范例。因此，我们认为，对于分布式应用程序的构建，DSM范例将比显式通讯（例如RPC，CORBA）更具吸引力。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号