Scholarlink

<p>In distributed systems with redundant network paths, dynamically selecting optimal communication routes for parallel programs is essential for minimizing latency and avoiding congestion. However, this is challenging due to unpredictable network conditions and concurrent workloads that create time-varying performance characteristics. This paper presents a reinforcement learning framework that enables programs to adaptively select communication routes based on historical performance without requiring global network state monitoring. We employ the Upper Confidence Bound 1 (UCB1) algorithm for small candidate sets and an improved <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\epsilon \)</EquationSource> </InlineEquation>-greedy algorithm for larger sets, providing logarithmic regret bounds in stationary environments and sublinear regret in dynamic scenarios. We demonstrate this approach on the 3-Quads cluster, a distributed system with three redundant subnetworks, where simulation and visualization programs run concurrently. Experiments show that our method reduces data transmission delay by 30-45% compared to random routing, with failure rates below 6% and overhead less than 0.6%. The approach converges to near-optimal routes within 500 communication rounds across different data sizes. While demonstrated experimentally on 3-Quads, theoretical analysis indicates that the framework can generalize to other redundant network topologies including fat-tree and dragonfly networks, with performance guarantees whose dependence on network topology is limited to the candidate set size <i>m</i>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dynamic Communication Optimization with Collision Avoidance for Parallel Programs in Distributed Systems

  • 期刊论文

  • 发表日期：

    2026年3月17日

查看全文
  • Zhoukai Wang,
  • Shin-ichiro Mori

摘要

In distributed systems with redundant network paths, dynamically selecting optimal communication routes for parallel programs is essential for minimizing latency and avoiding congestion. However, this is challenging due to unpredictable network conditions and concurrent workloads that create time-varying performance characteristics. This paper presents a reinforcement learning framework that enables programs to adaptively select communication routes based on historical performance without requiring global network state monitoring. We employ the Upper Confidence Bound 1 (UCB1) algorithm for small candidate sets and an improved \(\epsilon \) -greedy algorithm for larger sets, providing logarithmic regret bounds in stationary environments and sublinear regret in dynamic scenarios. We demonstrate this approach on the 3-Quads cluster, a distributed system with three redundant subnetworks, where simulation and visualization programs run concurrently. Experiments show that our method reduces data transmission delay by 30-45% compared to random routing, with failure rates below 6% and overhead less than 0.6%. The approach converges to near-optimal routes within 500 communication rounds across different data sizes. While demonstrated experimentally on 3-Quads, theoretical analysis indicates that the framework can generalize to other redundant network topologies including fat-tree and dragonfly networks, with performance guarantees whose dependence on network topology is limited to the candidate set size m.