Improving LSM‐trie performance by parallel search

Abstract

Summary LSM‐trie‐based key‐value (KV) store is often used to manage an ultralarge dataset in reality by introducing a number of sublevels at each level, its linear growth pattern can fairly reduce the write amplification in store operations. Although this design is effective for the write operation, the last level holds a large proportion of KV items, leading to the extreme imbalance of data distribution. Therefore, to support efficient read, we need to carefully consider this imbalance. On the other hand, to ensure that acquired data is latest, the LSM‐trie needs to search the dataset at different levels one by one, and this search method may take a lot of unnecessary time. When the number of items is ultralarge, the random lookup performance may be poor due to the imbalance data distribution. To address this issue, we improve the read performance of the LSM‐trie by changing its serial search to parallel search, using two threads to simultaneously search at the last level and other levels, respectively. Our experiment results show that the read performance of the LSM‐trie can be improved up to 98.35% and on average 71.55%.

An LSM-Tree Index for Spatial Data

“…It introduces the update memo into the LSM R-tree index, reduces the complexity of index deletion and update and improves the performance of data queries. The problem of reading amplification can be partially solved in traditional ways to improve the performance of LSM-tree reading, such as an LSM-tree tidal structure [14] and a multi-threaded parallel query [15]. In the tidal structure, Wang et al moved the files frequently visited in the bottom to a higher position, reducing the data reading delay and thus reducing the system read amplification.…”

“…Such dual index querying leads to significant reading amplification. Although the read amplification problem can be alleviated by improving the LSM read performance, such as the LSM-tree tidal structure [14] and multi-threaded parallel querying [15], it still does not fundamentally solve the mechanical problem of a secondary index query [16]. An R-tree index is the most widely used spatial index among tree-like auxiliary indexes and its query performance is stable and applicable to a wide range of data types.…”

SPOPB: Reducing solid state drive write traffic for flash‐based key‐value caching

“…By hashing the key and deterministically mapping the value to storage space, the GC overhead can be significantly mitigated. To improve the problem of inefficient read of the LSM‐tree‐based KV store, Cheng's 26 work proposes a method of parallel search by carefully considering the imbalance of data distribution in the last level of the LSM‐tree structure. Kassa's 27 work focuses on the read‐dominant workloads with strong locality, and introduces Storage Class Memories (SCM) as a volatile memory extension.…”