An efficient design and implementation of LSM-tree based key-value store on open-channel SSD

Abstract

Various key-value (KV) stores are widely employed for data management to support Internet services as they offer higher efficiency, scalability, and availability than relational database systems. The log-structured merge tree (LSM-tree) based KV stores have attracted growing attention because they can eliminate random writes and maintain acceptable read performance. Recently, as the price per unit capacity of NAND flash decreases, solid state disks (SSDs) have been extensively adopted in enterprise-scale data centers to provide high I/O bandwidth and low access latency. However, it is inefficient to naively combine LSM-tree-based KV stores with SSDs, as the high parallelism enabled within the SSD cannot be fully exploited. Current LSM-tree-based KV stores are designed without assuming SSD's multi-channel architecture.To address this inadequacy, we propose LOCS, a system equipped with a customized SSD design, which exposes its internal flash channels to applications, to work with the LSM-tree-based KV store, specifically LevelDB in this work. We extend LevelDB to explicitly leverage the multiple channels of an SSD to exploit its abundant parallelism. In * This work was done during their visiting professorship at the Center for Energy-Efficient Computing and Applications in Peking University.† Jason Cong is also a co-director of the PKU/UCLA Joint Research Institute in Science and Engineering.Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from permissions@acm.org. addition, we optimize scheduling and dispatching polices for concurrent I/O requests to further improve the efficiency of data access. Compared with the scenario where a stock LevelDB runs on a conventional SSD, the throughput of storage system can be improved by more than 4× after applying all proposed optimization techniques.

LSM-based storage techniques: a survey

“…In this subsection, we have reviewed the LSM-tree improvements exploiting hardware platforms, including large memory [15,19], multi-core [34], SSD/NVM [13,20,39,44,46,54,71], and native storage [49,74,73]. To manage large memory components, both FloDB [15] and Accordion [19] take a multi-layer approach to limit random writes to a small memory area.…”

“…Meanwhile, values must be garbagecollected separately to reclaim disk space, which is similar to the traditional log-structured file system design [60]. Finally, some recent work has proposed to perform native management of storage devices, including HDDs [49] and SSDs [73,74], which can often bring large performance gains by exploiting the sequential and non-overwriting I/O patterns exhibited by LSM-trees.…”

“…In contrast, dCompaction [53] offers no builtin support for workload balancing. It is not clear how skewed SSTable groups would impact the performance of these struc- ♣ LWC-tree [78,79] ♣ PebblesDB [58] ♣ dCompaction [53] ♣ Zhang et al [82] ♣ SifrDB [50] ♣ Skip-tree [81] ♣ TRIAD [16] ♣ VT-tree [64] ♣ Zhang et al [84] ♣ Ahmad et al [8] ♣ LSbM-tree [68,69] ♣ bLSM [61] ♣ FloDB [15] ♣ Accordion [19] ♣ cLSM [34] ♣ FD-tree [44] ♣ FD+tree [71] ♣ MaSM [13] ♣ WiscKey [46] ♣ HashKV [20] ♣ Kreon [54] ♣ NoveLSM [39] ♣ LDS [49] ♣ LOCS [74] ♣ NoFTL-KV [73] ♣ LHAM [51] ♣ LSM-trie [76] ♣ SlimDB [59] ♣ Mathieu et al [48] ♣ Lim et al [45] ♣ Monkey [25,26] ♣ Dostoevsky [24] ♣ Thonangi and Yang [70] ♣ ElasticBF [83] ♣ Mutant [80] ♣ LSII [75] ♣ Kim et al [42] ♣ Filter [11] ♣ Qader et al [57] ♣ Diff-Index [66] ♣ DELI [67], ♣ Luo and Carey …”

“…LOCS [74] is Fig. 13: WiscKey stores values into an append-only log to reduce the write amplification of the LSM-tree allows applications to flexibly schedule disk writes to leverage the available I/O parallelism, but disk reads must be served by the same channel where the data is stored.…”

Efficient Tree Indexing for PCM-Based Memory Systems

“…In recent years, the research of B+-tree index based on flash memory has been fruitful [10,11], but PCM-based research is still rare. Due to the different characteristics flash and PCM (such as flash memory is ectopic update, block erase, and PCM can be updated in situ, byte addressable, etc.…”