Scholarlink

<p>Key-value (KV) stores are widely utilized in data-intensive applications to obtain exceptional storage performance. However, its caching mechanism often suffers read and write pauses. Especially when accessing old data periodically, it results in cache hit ratios and system throughput decline. To address the performance degradation issue, we propose an innovative dual-hash caching mechanism called DHCache. Firstly, we introduce a dual-hash structure in DHCache. It alleviates read and write pauses by reducing the frequency of rehash operations on the hash table. Secondly, we employ a Most Recently Used (MRU) cache replacement policy on DHCache to retain old data. This enhances the cache hit ratios and throughput when periodically accessing old data. DHCache is deployed within LevelDB, demonstrating significant performance advantages. Experimental results indicate that DHCache improves throughput by 11.89–21.92% in various read workloads compared to traditional LRUCache. Significantly, read performance improvement does not come at the cost of write performance degradation.</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

Dhcache: a dual-hash cache for optimizing the read performance in key-value store

  • 期刊论文

  • 发表日期：

    2025年1月19日

查看全文
  • Jinkang Lu,
  • Meng Lv,
  • Peixuan Li,
  • Zhu Yuan,
  • Ping Xie

摘要

Key-value (KV) stores are widely utilized in data-intensive applications to obtain exceptional storage performance. However, its caching mechanism often suffers read and write pauses. Especially when accessing old data periodically, it results in cache hit ratios and system throughput decline. To address the performance degradation issue, we propose an innovative dual-hash caching mechanism called DHCache. Firstly, we introduce a dual-hash structure in DHCache. It alleviates read and write pauses by reducing the frequency of rehash operations on the hash table. Secondly, we employ a Most Recently Used (MRU) cache replacement policy on DHCache to retain old data. This enhances the cache hit ratios and throughput when periodically accessing old data. DHCache is deployed within LevelDB, demonstrating significant performance advantages. Experimental results indicate that DHCache improves throughput by 11.89–21.92% in various read workloads compared to traditional LRUCache. Significantly, read performance improvement does not come at the cost of write performance degradation.