Scholarlink

<p>Modern tiered data storage systems combine different media device types&#xa0;(e.g., HDD, SSD, NVRAM) with several mechanisms (such as caching, tiering, data distribution, and fault tolerance protocols) distributed across storage nodes to optimize data access. Evaluating the impact of updates on the storage setup or changes in such mechanisms can be time-consuming, often requiring system modification and redeployment. To address this challenge, we present DITIS, a new system-level simulator for distributed tiered data storage systems that captures the end-to-end execution path of file system requests across storage nodes and layers, using numerous pluggable policies that control key aspects of execution. DITIS can be configured with varying numbers of storage tiers, caches, nodes, and media devices, each modeled with distinct performance characteristics through pluggable performance cost models. DITIS’ architecture is based on an adaptation of the Actor Model, where each key component exchanges synchronous or asynchronous messages, mirroring a distributed multi-threaded system. DITIS also includes an optimizer for exploring configuration parameter spaces and identifying optimal setups for specific objectives. DITIS is evaluated using production workload traces from Huawei Technologies and SNIA IOTTA. The results show that DITIS can accurately simulate millions of file I/O requests over gigabytes of data within seconds, offering developers an efficient tool for navigating the complex design space of modern tiered storage systems.</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

DITIS: An End-to-End System-Level Simulator and Optimizer for Distributed Tiered Storage

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月16日

查看全文
  • Edson Ramiro Lucas Filho,
  • Andreas Efstathiou,
  • Lun Yang,
  • Kebo Fu,
  • Jianqiang Shen,
  • Herodotos Herodotou

摘要

Modern tiered data storage systems combine different media device types (e.g., HDD, SSD, NVRAM) with several mechanisms (such as caching, tiering, data distribution, and fault tolerance protocols) distributed across storage nodes to optimize data access. Evaluating the impact of updates on the storage setup or changes in such mechanisms can be time-consuming, often requiring system modification and redeployment. To address this challenge, we present DITIS, a new system-level simulator for distributed tiered data storage systems that captures the end-to-end execution path of file system requests across storage nodes and layers, using numerous pluggable policies that control key aspects of execution. DITIS can be configured with varying numbers of storage tiers, caches, nodes, and media devices, each modeled with distinct performance characteristics through pluggable performance cost models. DITIS’ architecture is based on an adaptation of the Actor Model, where each key component exchanges synchronous or asynchronous messages, mirroring a distributed multi-threaded system. DITIS also includes an optimizer for exploring configuration parameter spaces and identifying optimal setups for specific objectives. DITIS is evaluated using production workload traces from Huawei Technologies and SNIA IOTTA. The results show that DITIS can accurately simulate millions of file I/O requests over gigabytes of data within seconds, offering developers an efficient tool for navigating the complex design space of modern tiered storage systems.