Key-value Store Research Articles

TrieKV: A High-Performance Key-Value Store Design With Memory as Its First-Class Citizen

TrieKV: A High-Performance Key-Value Store Design With Memory as Its First-Class Citizen

SuccinctKV: a CPU-efficient LSM-tree Based KV Store with Scan-based Compaction

The CPU overhead of the LSM-tree becomes increasingly significant when high-speed storage devices are utilized. In this article, we propose SuccinctKV , a key-value store based on LSM-tree that is optimized to improve CPU efficiency in mixed workload scenarios. To achieve this, SuccinctKV reduces the CPU overhead of compaction by writing scan-sorted data directly to the storage device. SuccinctKV also redesigns the merge-sort operation of the LSM-tree, enhancing CPU locality and reducing the unnecessary CPU overhead of cache accesses and I/O system calls. Additionally, SuccinctKV introduces a scheduler to resolve potential bursty I/O contention by autonomously initiating I/O requests at appropriate times and quickly relieving I/O pressure by terminating background I/O requests. We implement SuccinctKV on RocksDB and conduct extensive experiments to evaluate our proposed methods. The experimental results demonstrate that, compared with RocksDB, SuccinctKV achieves a maximum improvement of 2.6x in scan performance and reduces CPU overhead of compaction by up to 89% under mixed workloads.

Time-constrained persistent deletion for key–value store engine on ZNS SSD

The inherent out-of-place update characteristic of the Log-Structured Merge tree (LSM tree) cannot guarantee persistent deletion within a specific time window, leading to potential data privacy and security issues. Existing solutions like Lethe-Fade ensure time-constrained persistent deletion but introduce considerable write overhead, worsening the write amplification issue, particularly for key–value stores on ZNS SSD. To address this problem, we propose a zone-aware persistent deletion scheme for key–value store engines. Targeting mitigating the write amplification induced by level compaction, we design an adaptive SSTable selection strategy for each level in the LSM tree. Additionally, as the SSTable with deletion records would become invalid after the persistent deletion timer reaches its threshold, we design a tombstone-aware zone allocation strategy to reduce the data migration induced by garbage collection. In further, we optimize the victim zone selection in GC to reduce the invalid migration of tombstone files. Experimental results demonstrate that our scheme effectively ensures that most outdated physical versions are deleted before reaching the persistent deletion time threshold. When deleting 10% of keys in the key–value store engine, this scheme reduces write amplification by 74.7% and the garbage collection-induced write by 87.3% compared to the Lethe-Fade scheme.

IFKVS: Lightweight Key-Value Store for Flash-Based Intermittently Computing Devices

iFKVS: Lightweight Key-Value Store for Flash-Based Intermittently Computing Devices

EKV-VBQ: Ensuring Verifiable Boolean Queries in Encrypted Key-Value Stores.

To address the deficiencies in privacy-preserving expressive query and verification mechanisms in outsourced key-value stores, we propose EKV-VBQ, a scheme designed to ensure verifiable Boolean queries over encrypted key-value data. We have integrated blockchain and homomorphic Xor operations and pseudo-random functions to create a secure and verifiable datastore, while enabling efficient encrypted Boolean queries. Additionally, we have designed a lightweight verification protocol using bilinear map accumulators to guarantee the correctness of Boolean query results. Our security analysis demonstrates that EKV-VBQ is secure against adaptive chosen label attacks (IND-CLA) and guarantees Integrity and Unforgeability under the bilinear q-strong Diffie-Hellman assumption. Our performance evaluations showed reduced server-side storage overhead, efficient proof generation, and a significant reduction in user-side computational complexity by a factor of log n. Finally, GPU-accelerated optimizations significantly enhance EKV-VBQ's performance, reducing computational overhead by up to 50%, making EKV-VBQ highly efficient and suitable for deployment in environments with limited computational resources.

CDNRocks: computable data nodes with RocksDB to improve the read performance of LSM-tree-based distributed key-value storage systems

CDNRocks: computable data nodes with RocksDB to improve the read performance of LSM-tree-based distributed key-value storage systems

Scalable Metadata-Hiding for Privacy-Preserving IoT Systems

Modern cloud-based IoT services comprise an integrator service and several device vendor services. The vendor services enable users to remotely control their devices, while the integrator serves as a central intermediary, offering a unified interface for managing devices from different vendors. Although such a model is quite beneficial for IoT services to evolve quickly, it also creates a serious privacy concern: the vendor and integrator services observe all interactions between users and devices. Toward this, we propose Mohito, a privacy-preserving IoT system that hides such interactions from both the integrator and the vendors. In Mohito, we protect both the interaction data and the metadata, so that no one learns which user is communicating with which device. By utilizing oblivious key-value storage as a primitive and leveraging the unique communication graph of IoT services, we build a scalable protocol specialized in handling large concurrent traffic, a common demand in IoT systems. Our evaluation shows that Mohito can achieve up to 600x more throughput than the state-of-the-art general-purpose systems that provide similar security guarantees.

CAMAL: Optimizing LSM-trees via Active Learning

We use machine learning to optimize LSM-tree structure, aiming to reduce the cost of processing various read/write operations. We introduce a new approach CAMAL, which boasts the following features: (1) ML-Aided : CAMAL is the first attempt to apply active learning to tune LSM-tree based key-value stores. The learning process is coupled with traditional cost models to improve the training process; (2) Decoupled Active Learning : backed by rigorous analysis, CAMAL adopts active learning paradigm based on a decoupled tuning of each parameter, which further accelerates the learning process; (3) Easy Extrapolation : CAMAL adopts an effective mechanism to incrementally update the model with the growth of the data size; (4) Dynamic Mode : CAMAL is able to tune LSM-tree online under dynamically changing workloads; (5) Significant System Improvement : By integrating CAMAL into a full system RocksDB, the system performance improves by 28% on average and up to 8x compared to a state-of-the-art RocksDB design.

Performance of Linear and Spiral Hashing Algorithms

Linear Hashing is an important algorithm for many key-value stores in main memory. Spiral Storage was invented to overcome the poor fringe behavior of Linear Hashing, but after an influential study by Larson, it seems to have been discarded. Since almost 50 years have passed, we repeat Larson’s comparison with the in-memory implementation of both to see whether his verdict still stands. Our study shows that Spiral Storage has slightly better lookup performance but slightly poorer insert performance. However, Spiral Hashing has more predictable performance for inserts and, in particular, better fringe behavior.

Improving Performance of Key–Value Stores for High-Performance Storage Devices

Key–value stores (KV stores) are becoming popular in both academia and industry due to their high performance and simplicity in data management. Unlike traditional database systems such as relational databases, KV stores manage data in key–value pairs and do not support relationships between the data. This simplicity enables KV stores to offer higher performance. To further improve the performance of KV stores, high-performance storage devices such as solid-state drives (SSDs) and non-volatile memory express (NVMe) SSDs have been widely adopted. These devices are intended to expedite data processing and storage. However, our studies indicate that, due to a lack of multi-thread-oriented programming, the performance of KV stores is far below the raw performance of high-performance storage devices. In this paper, we analyze the performance of existing KV stores utilizing high-performance storage devices. Our analysis reveals that the actual performance of KV stores is below the potential performance that these storage devices could offer. According to the profiling results, we argue that this performance gap is due to the coarse-grained locking mechanisms of existing KV stores. To alleviate this issue, we propose a multi-threaded compaction operation that leverages idle threads to participate in I/O operations. Our experimental results demonstrate that our scheme can improve the performance of KV stores by up to 16% by increasing the number of threads involved in I/O operations.

Efficient Wear-Leveling-Aware Data Placement for LSM-Tree based key-value store on ZNS SSDs

Emerging Zoned Namespace (ZNS) is a new-style Solid State Drive (SSD) that manages data in a zoned manner, which can achieve higher performance by strictly obeying the sequential write mode in each zone and alleviating the redundant overhead of garbage collections. Unfortunately, flash memory usually has a serious problem with limited program/erase cycles. Meanwhile, inappropriate data placement strategy of storage systems can lead to imbalanced wear among zones, severely reducing the lifespan of ZNS SSDs. In this paper, we propose a Wear-Leveling-Aware Data Placement (WADP) to solve this problem with negligible performance cost. First, WADP employs a wear-aware empty zone allocation algorithm to quantify the resets of zones and choose the less-worn zone for each allocation. Second, to prevent long-term zone occupation of infrequently written data (namely cold data), we propose a wear-leveling cold zone monitoring mechanism to identify cold zones dynamically. Finally, WADP adopts a real-time I/O pressure-aware data migration mechanism to adaptively migrate cold data for achieving wear-leveling among zones. We implement the proposed WADP in ZenFS and evaluate it with widely used workloads. Compared with state-of-the-art solutions, i.e., LIZA and FAR, the experimental results show that WADP can significantly reduce the standard deviation of zone resets while maintaining decent performance.

LavaStore: ByteDance's Purpose-Built, High-Performance, Cost-Effective Local Storage Engine for Cloud Services

Persistent key-value (KV) stores are widely used by cloud services at ByteDance as local storage engines, and RocksDB used to be the de facto implementation since it can be tailored to a variety of workloads and requirements. In this paper, we provide key insights into local storage engine usage at ByteDance, explain why the combination of highly write-intensive workloads and stringent requirements on cost efficiency and point lookup tail latency may pose challenges to a general-purpose local storage engine such as RocksDB, and present the design and implementation of LavaStore , a high-performance cost-effective local storage engine purpose-built to address these challenges. LavaStore achieves its design goals by selectively customizing a few components of a RocksDB-based, general-purpose local storage engine, including a distinct KV separation design that decouples garbage collection from compaction, a specialized engine type for the commonly recurring Write-Ahead-Logging workload, and a customized user-space append-only filesystem. LavaStore has been deployed to production with hundreds of thousands of running instances, storing more than 100 PB of data and serving billions of requests per second, bringing significant performance improvements and cost reductions to customers over their original local storage engines. For example, a ByteDance proprietary distributed OLTP database service has experienced a reduction in average write and read latency by 61% and 16%, respectively, and a ByteDance proprietary caching service has gained an 87% increase in write throughput with no more than 6% space overhead.

Presto's History-Based Query Optimizer

An important feature of modern query optimizers is the ability to produce a query plan that is optimal for the underlying data set. This requires the ability to estimate cardinalities and computational costs of intermediate query plan nodes, which is highly dependent on both the query shape and the underlying data distribution. Traditional methods include collecting statistics on base tables and implementing cardinality and computational cost derivation inside the optimizer, which is error-prone for complex query shapes. This paper presents Presto's novel history-based optimization framework (HBO), which collects execution histories and uses them to optimize similar queries in the future. The framework produces accurate estimates for complex query shapes in a lightweight, automated manner, and adapts automatically to changes in underlying data distributions. We present the design and implementation of the HBO framework and provide details on its use in various optimization rules, as well as details on implementing the statistics store on top of a Redis key-value store. We also present the results of running HBO in production in two large data infrastructure organizations (Meta and Uber).

Efficient multi-party PSI and its application in port management

Private Set Intersection (PSI) technology is a cryptographic tool that allows the parties holding private data to determine the intersection of sets in a joint computation without revealing any additional privacy information. As a critical component of secure multi-party computation, this technology has been widely used in the security domain of artificial intelligence and data mining. With the emergence of the era of multi-source data sharing, protocols for private set intersection computation applicable to multiple participants have also emerged. However, the performance of existing multi-party private set intersection (MPSI) protocols is suboptimal when some participants use devices with limited communication or computational capabilities, such as mobile devices. To overcome the above issues, we design a cloud-aided multi-party private set intersection protocol (Cloud-Aided-MPSI) based on oblivious programmable pseudorandom functions (OPPRF) and oblivious key–value stores (OKVS). Due to the protocol efficiently outsourcing partial computational and communication tasks to cloud servers, our performance has been further enhanced compared to existing work. Through the Cloud-Aided-MPSI protocol, we propose a port scheduling protocol for coordinated management scenarios of ships arriving and departing from ports. The protocol effectively addresses the privacy protection concerns of port management when scheduling the arrival and departure of ships. We analyze the performance of this protocol.

Privacy-Preserving Membership Queries for Federated Anomaly Detection

In this work, we propose a new privacy-preserving membership query protocol that lets a centralized entity privately query datasets held by one or more other parties to check if they contain a given element. This protocol, based on elliptic curve-based ElGamal and oblivious key-value stores, ensures that those 'data-augmenting' parties only have to send their encrypted data to the centralized entity once, making the protocol particularly efficient when the centralized entity repeatedly queries the same sets of data. We apply this protocol to detect anomalies in cross-silo federations. Data anomalies across such cross-silo federations are challenging to detect because (1) the centralized entities have little knowledge of the actual users, (2) the data-augmenting entities do not have a global view of the system, and (3) privacy concerns and regulations prevent pooling all the data. Our protocol allows for anomaly detection even in strongly separated distributed systems while protecting users' privacy. Specifically, we propose a cross-silo federated architecture in which a centralized entity (the backbone) has labeled data to train a machine learning model for detecting anomalous instances. The other entities in the federation are data-augmenting clients (the user-facing entities) who collaborate with the centralized entity to extract feature values to improve the utility of the model. These feature values are computed using our privacy-preserving membership query protocol. The model can be trained with an off-the-shelf machine learning algorithm that provides differential privacy to prevent it from memorizing instances from the training data, thereby providing output privacy. However, it is not straightforward to also efficiently provide input privacy, which ensures that none of the entities in the federation ever see the data of other entities in an unencrypted form. We demonstrate the effectiveness of our approach in the financial domain, motivated by the PETs Prize Challenge, which is a collaborative effort between the US and UK governments to combat international fraudulent transactions. We show that the private queries significantly increase the precision and recall of the otherwise centralized system and argue that this improvement translates to other use cases as well.

Towards Application Centric Carbon Emission Management

Carbon emissions are due to application execution on a target system (operational emissions) and the production, transportation, and disposal of the system itself (embodied emissions). This paper investigates the impacts of different resource configurations in terms of available DRAM memory on the overall carbon emission of individual application executions. We first propose an application-centric carbon footprint model that considers DRAM and CPU. We then study the model using a widely-used key-value store (RocksDB) and Graph500 applications. The results for RocksDB indicate that the minimal emission configuration is application dependent and can lead to significant emission reductions compared to application-oblivious configurations that use higher DRAM capacity without improving performance. For some applications, small performance degradation may lead to substantial additional emission reductions.

AStore: Uniformed Adaptive Learned Index and Cache for RDMA-Enabled Key-Value Store

AStore: Uniformed Adaptive Learned Index and Cache for RDMA-Enabled Key-Value Store

Index Shipping for Efficient Replication in LSM Key-Value Stores with Hybrid KV Placement

Key-value (KV) stores based on the LSM tree have become a foundational layer in the storage stack of datacenters and cloud services. Current approaches for achieving reliability and availability favor reducing network traffic and send to replicas only new KV pairs. As a result, they perform costly compactions to reorganize data in both the primary and backup nodes, which increases device I/O traffic and CPU overhead, and eventually hurts overall system performance. In this article, we describe Tebis , an efficient LSM-based KV store that reduces I/O amplification and CPU overhead for maintaining the replica index. We use a primary-backup replication scheme that performs compactions only on the primary nodes and sends pre-built indexes to backup nodes, avoiding all compactions in backup nodes. Our approach includes an efficient mechanism to deal with pointer translation across nodes in the pre-built region index. Our results show that Tebis reduces resource utilization on backup nodes compared to performing full compactions: throughput is increased by 1.06 to 2.90×, CPU efficiency is increased by 1.21 to 2.78×, and I/O amplification is reduced by 1.7 to 3.27×, whereas network traffic increases by up to 1.32 to 3.76x.

A Semantic-Integrated LSM-Tree-Based Key–Value Storage Engine for Blockchain Systems

A Semantic-Integrated LSM-Tree-Based Key–Value Storage Engine for Blockchain Systems

A Fast Learned Key-Value Store for Concurrent and Distributed Systems

A Fast Learned Key-Value Store for Concurrent and Distributed Systems