【24h】

Reducing Read, Write Latencies in Distributed Objects Through Replication and Caching

机译:通过复制和缓存减少分布式对象中的读写延迟

获取原文
获取原文并翻译 | 示例

摘要

Read and write operations in distributed objects have traditionally been exposed to network latencies. This results in such invocations being expensive in cases where network latencies are high. Temporal locality has been exploited in several products through the use of object mobility to achieve read and write time reduction. But such an approach hinges latency reduction for both reads and writes on temporal locality. We studied the replication and caching strategy employed by several distributed shared memories and implemented it in CORBA, a contemporary distributed object system. Together with a variant of relaxed consistency property, which ensures the correctness of program execution in the presence of object replicas, the strategy eliminates the dependency of read latency reduction on temporal locality. It also shields read times from the effect of frequent remote writes, each of which triggers a local state update. Write latency reduction is dependent on the property of temporal locality. But results obtained show that a small number of nodes, coupled with a small (invocation intervaletwork latency) ratio, could produce the same effect in the absence of the prop- erty. Further, a small ratio of invocation intervals also helps reduce write latencies experienced by nodes with shorter intervals.
机译:传统上,分布式对象中的读取和写入操作已暴露于网络延迟中。这导致在网络等待时间较长的情况下,此类调用非常昂贵。通过使用对象移动性,已在几种产品中开发了时间局部性,以减少读写时间。但是,这种方法取决于在时间局部性上减少读写的等待时间。我们研究了多个分布式共享内存所采用的复制和缓存策略,并在现代分布式对象系统CORBA中实现了该策略。该策略与宽松一致性属性的变体一起使用,可以确保存在对象副本时程序执行的正确性,该策略消除了读取等待时间减少对时间局部性的依赖性。它还可以保护读取时间免受频繁远程写入的影响,每次远程写入都会触发本地状态更新。减少写等待时间取决于时间局部性的属性。但是获得的结果表明,在没有该属性的情况下,少量节点加上较小的(调用间隔/网络等待时间)比率可以产生相同的效果。此外,较小的调用间隔比率也有助于减少间隔较短的节点所经历的写延迟。

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