The development of data analytics services has fueled many optimizations in data scans, and indexes are one of the most important techniques to improve scan efficiency. Meanwhile, block-based data organization has become standard practice in these services, providing an opportunity for more fine-grained index selection at the block level. However, today’s systems ignore data distribution differences among blocks and usually tune indexes over the entire database table, leading to unnecessary storage costs and potential degradation in query performance. To bridge this gap, we propose BISLearner, a fast, block-aware index selecting approach based on reinforcement learning. One major challenge lies in differentiating the data distribution among data blocks. To solve this problem, BISLearner maintains simplified histograms that represent the data distribution of each block. When a query is issued, BISLearner leverages the query predicate and histogram-based block summaries to generate a specific workload representation for each block. However, such block-aware workload representation leads to an excessive number of input features, resulting in a slow or even incorrect convergence of neural networks. Inspired by the human-learning process, where more attention is devoted to the important parts of data, we design an attention-based neural model to efficiently handle the high volume of input features caused by table partitioning and select the best-suited index combinations at the block level. Additionally, to handle the expensive search space caused by attribute combinations and data partitioning, we employ heuristic-based invalid action masking at the block level to accelerate the training process. Our evaluation using PostgreSQL and Greenplum database systems demonstrates BISLearner is able to reduce job completion time by up to 28.45% compared to its best counterparts.

Using a machine learning (ML) model as a core component in database knob tuning has demonstrated remarkable advancements in recent years. However, a model that optimizes both categorical and continuous values in the same way may not guarantee efficiency and effectiveness in knob tuning. This is due to the fact that the usual assumption of a differentiable input space for efficient exploration of continuous spaces does not hold true in categorical spaces. Moreover, the inherent complexity of interdependences among knobs and the high-dimensionality of the configuration space compound the challenges of tuning. In this paper, we propose TopTune, which employs tailored optimization for continuous and categorical knobs, to achieve accelerated tuning efficiency and improved tuning performance. Specifically, we decompose the configuration space into two orthogonal subspaces: categorical and continuous spaces. Subsequently, we employ Bayesian optimization models, i.e., SMAC and GP to explore the categorical and continuous subspaces, respectively. These two models will alternately explore the two spaces with the proposed communication mechanism to ensure TopTune can capture the dependence between continuous and categorical knobs. Furthermore, to balance efficiency and accuracy, we utilize a knob-dimensional projection strategy to reduce the exploration domain by embedding the high-dimension configuration space into a lower-dimensional proxy space. In addition, we implement batch Bayesian optimization technology, which enables parallel knob evaluation while balancing exploration and exploitation. We evaluate TopTune under different benchmarks (SYSBENCH, TPC-C, and JOB), metrics (throughput and latency), and DBMSs (MySQL and Dameng). Extensive experiments demonstrate that TopTune identifies better configurations in up to approximately 12.2× less time while achieving a 10.7% improvement in throughput compared to state-of-the-art methods.

The shattered and overlapped Level-0 data organization is the primary cause of write stall and read amplification problems in LSM-Tree-based Key-Value (KV) stores: (1) Level-0 to Level-1 compaction involves a large amount of data which induces write stalls, and (2) A point lookup needs to access multiple files in Level-0 which leads to significant read amplification. To address the problem, we propose BL-Tree by replacing the shattered Level-0 in LSM-Tree with a B+-Tree in byte-addressable Persistent Memory (PM). The sorted B+-Tree of Level-0 can accelerate the point lookup speed and reduce read/write amplification. BL-Tree further conducts the locality-aware and parallel compaction from the B+-Tree in PM (Level-0) to the lower levels of LSM-Tree in SSDs by only moving cold data downward, thus alleviating the write stalls and reducing the read/write amplification simultaneously. The extensive experiments on the prototype of BL- Tree show that it definitely avoids the write stalls and significantly reduces the read/write amplification. As a result, BL-Tree reduces the P99 tail latency by 65.2 × than LevelDB-PM and speeds up the throughput by more than 2 × under workloads with spatial locality than other KV stores.

Abstract Hashing-based unsupervised cross-modal retrieval has gained significant attention in the big data management community due to its low storage overhead and rapid retrieval speed. However, current methods often lack effective alignment strategies to reduce the modality gap. They also fail to explore the latent structural information of the training data for accurate relationship learning, resulting in sub-optimal cross-modal retrieval performance. To tackle these challenges, we propose a novel unsupervised cross-modal hashing method called G raph C ontrastive-and- R econstructive H ashing ( GCRH ). Specifically, GCRH first performs global graph contrastive learning , which involves both intra-modal and inter-modal pairs. This facilitates the learning of more discriminative hash codes through intra-modal discrimination and inter-modal alignment objectives. To further bridge the modality gap, GCRH conducts local graph reconstruction using GCN-based decoders to reconstruct the original features of one modality from the hash codes of another. The integration of contrastive-and-reconstructive learning with graph structural information enables GCRH to generate high-quality hash codes that are both well-aligned and discriminative. Extensive experiments on three benchmark datasets substantiate the superior cross-modal retrieval performance of GCRH .

Workload execution can account for 90% of the total database knob tuning time, which is often the bottleneck for efficient knob tuning in practice. Reducing the tuning time by using a compressed workload is a natural solution. However, many existing workload compression methods are designed for OLAP workloads, which reduce the number of queries needed for analysis tasks by sampling a small subset of queries. These methods are less effective for OLTP workloads in knob-tuning tasks, as they often disregard essential contextual details, including query sequence and concurrency. As a result, configurations that perform well on the compressed OLTP workload may not deliver similar competitive performance on the original workload. To address these challenges, we first define the objective of OLTP workload compression for knob tuning. We then propose a slice-based compression method, SCompression , which compresses workloads by slicing based on time intervals while preserving concurrency. SCompression achieves the objective by focusing on generating a compressed workload that (1) executes faster than the original workload and (2) produces performance variations similar to the source workload under different configurations. SCompression works in three steps: (1) dividing the workload into segments to capture regular performance fluctuations, (2) slicing each segment to preserve concurrency and transaction context, and (3) sampling slices under execution time constraints using a cluster-based approach to ensure representativeness. Finally, SCompression replays the compressed workload to produce the performance that mirrors the source workload. Extensive experiments on real-world and benchmark OLTP workloads show that SCompression is a cost-effective solution for knob tuning, accelerating tuning by up to 40× with only a 5% performance reduction.

In recent years at Tencent, we have observed that the use of multiple NoSQL databases for storing business data with diverse models has led to increased programming and deployment costs, as well as inefficient maintenance and underutilized resources. In this paper, we report X-Stor, a cloud-native NoSQL database system that supports multiple data models by extending different storage engines and efficiently managing them through a unified control plane. This design significantly reduces expenses and enables rapid expansion of new models, while seamlessly supporting their complete functionality through storage engine extensions. By consolidating multi-tenant services and data models on the same physical machines, X-Stor significantly enhances the utilization of cluster resources. Additionally, X-Stor introduces a standardized metric called Request Unit (RU) to measure tenant resource consumption for consumption-based pricing purposes. Leveraging this metric, we design RU-based resource management strategies and achieve efficient multi-tenant resource isolation and system load balancing. Currently, X-Stor manages a storage capacity of over 12PB for online operational data, including more than 100,000 tables with multiple data models. It handles 700 billion requests per day with a peak of 30 million requests per second. We evaluate the performance of X-Stor on popular benchmarks and production workloads. The results show that X-Stor performs well under diverse data models.

Meta-learning excels in few-shot learning by extracting shared knowledge from the observed tasks. However, it needs the tasks to adhere to the i.i.d. constraint, which is challenging to achieve due to complex task relationships between data content. Current methods that create tasks in a one-dimensional structure and use meta-learning to learn all tasks flatly struggle with extracting shared knowledge from tasks with overlapping concepts. To address this issue, we propose further constructing tasks from the same environment into hyper-tasks. Since the distributions of hyper-tasks and tasks in a hyper-task can both be approximated as i.i.d. due to further summarization, the meta-learning algorithm can capture shared knowledge more efficiently. Based on the hyper-task, we propose a hierarchical meta-learning paradigm to meta-learn the meta-learning algorithm. The paradigm builds a customized meta-learner for each hyper-task, which makes meta-learners more flexible and expressive. We apply the paradigm to three classic meta-learning algorithms and conduct extensive experiments on public datasets, which confirm the superiority of hierarchical meta-learning in the few-shot learning setting. The code is released at https://github.com/tuantuange/H-meta-learning.

Anomaly detection system plays an important role in maintaining the stability of cloud database. Existing studies mainly focus on significant deviations in multivariate time series, such as a combination of CPU utilization, transactions per second, etc, to detect abnormal issues. Due to the complexity of cloud database structure and functions, these approaches are difficult to achieve a balance among detection performance, detection efficiency and workload adaptability. In this paper, we propose DBCatcher, a cloud database online anomaly detection system based on indicator correlation. Through extensive analysis of real-world cloud database time series, we find the correlations among trends in the same key performance indicators across databases within the same unit, which inspires us to explore a time series correlation measurement method that can efficiently detect abnormal issues. Meanwhile, we design a flexible time window observation mechanism and an adaptive threshold learning policy to minimize misjudgment caused by key performance indicator fluctuations, greatly enhancing the detection performance and workload adaptability. We conduct extensive experiments under real-world and synthetic workloads. Experimental results show that DBCatcher significantly improves the detection performance and detection efficiency compared to existing methods.

Modern data analytics services are coupled with external data storage services, making I/O from remote cloud storage one of the dominant costs for query processing. Techniques such as columnar block-based data organization and compression have become standard practices for these services to save storage and processing cost. However, the problem of effectively skipping irrelevant blocks at low overhead is still open. Existing data-skipping efforts maintain lightweight summaries (e.g., min/max, histograms) for each block to filter irrelevant data. However, such techniques ignore patterns in real-world data, enabling ineffective use of the storage budget and may cause serious false positives. This paper presents Sieve, a learning-enhanced index designed to efficiently filter out irrelevant blocks by capturing data patterns. Specifically, Sieve utilizes piece-wise linear functions to capture block distribution trends over the key space. Based on the captured trends, Sieve trades off storage consumption and false positives by grouping neighboring keys with similar block distributions into a single region. We have evaluated Sieve using Presto, and experiments on real-world datasets demonstrate that Sieve achieves up to 80% reduction in blocks accessed and 42% reduction in query times compared to its counterparts.

Recently, using machine learning for performance tuning of cloud database (CDB) service has shown great potentials. However, facing personalized requirements such as various restrictions for tuning with very different workloads, pre-trained models may mismatch or recommend suboptimal configurations given a new workload. On the other hand, if the system tunes configurations in an online fashion, the system will suffer from the cold start problem, resulting in long tuning time and performance fluctuation. To accommodate these problems, we propose an online CDB tuning system called HUNTER. The key feature of HUNTER is a hybrid architecture, which uses samples generated by Genetic Algorithm to warm-start the finer grained exploration of deep reinforcement learning. Meanwhile, we employ Principal Component Analysis, Random Forest, and Fast Exploration Strategy to reduce the search space and the update time of the learning model. In addition, we further propose a clone and parallelization scheme to stress-test workloads on multiple cloned CDB instances (CDBs), resulting in faster and safer configuration exploration. Extensive trials on CDB with public and real-world workloads demonstrate that, given the same time budget and resources, HUNTER improves performance and considerably decreases recommendation time compared to state-of-the-art tuning systems, with accelerations of up to 2.8× and 22.8× utilizing 1 and 20 cloned CDBs, respectively.

Abstract Configuration tuning is vital to optimize the performance of a database management system (DBMS). It becomes more tedious and urgent for cloud databases (CDB) due to diverse database instances and query workloads, which make the job of a database administrator (DBA) very difficult. Existing solutions for automatic DBMS configuration tuning have several limitations. Firstly, they adopt a pipelined learning model but cannot optimize the overall performance in an end-to-end manner. Secondly, they rely on large-scale high-quality training samples which are hard to obtain. Thirdly, existing approaches cannot recommend reasonable configurations for a large number of knobs to tune whose potential values live in such high-dimensional continuous space. Lastly, in cloud environments, existing approaches can hardly cope with the changes of hardware configurations and workloads, and have poor adaptability. To address these challenges, we design an end-to-end automatic CDB tuning system, $${\texttt {CDBTune}}^{+}$$ CDBTune + , using deep reinforcement learning (RL). $${\texttt {CDBTune}}^{+}$$ CDBTune + utilizes the deep deterministic policy gradient method to find the optimal configurations in a high-dimensional continuous space. $${\texttt {CDBTune}}^{+}$$ CDBTune + adopts a trial-and-error strategy to learn knob settings with a limited number of samples to accomplish the initial training, which alleviates the necessity of collecting a massive amount of high-quality samples. $${\texttt {CDBTune}}^{+}$$ CDBTune + adopts the reward-feedback mechanism in RL instead of traditional regression, which enables end-to-end learning and accelerates the convergence speed of our model and improves the efficiency of online tuning. Besides, we propose effective techniques to improve the training and tuning efficiency of $${\texttt {CDBTune}}^{+}$$ CDBTune + for practical usage in a cloud environment. We conducted extensive experiments under 7 different workloads on real cloud databases to evaluate $${\texttt {CDBTune}}^{+}$$ CDBTune + . Experimental results showed that $${\texttt {CDBTune}}^{+}$$ CDBTune + adapts well to a new hardware environment or workload, and significantly outperformed the state-of-the-art tuning tools and DBA experts.

In multi-label image recognition, it has become a popular method to predict those labels that co-occur in an image via modeling the label dependencies. Previous works focus on capturing the correlation between labels, but neglect to effectively fuse the image features and label embeddings, which severely affects the convergence efficiency of the model and inhibits the further precision improvement of multi-label image recognition. To overcome this shortcoming, in this paper, we introduce Multi-modal Factorized Bilinear pooling (MFB) which works as an efficient component to fuse cross-modal embeddings and propose F-GCN, a fast graph convolution network (GCN) based multi-label image recognition model. F-GCN consists of three key modules: (1) an image representation learning module which adopts a convolution neural network (CNN) to learn and generate image representations, (2) a label co-occurrence embedding module which first obtains the label vectors via the word embeddings technique and then adopts GCN to capture label co-occurrence embeddings and (3) an MFB fusion module which efficiently fuses these cross-modal vectors to enable an end-to-end model with a multi-label loss function. We conduct extensive experiments on two multi-label datasets including MS-COCO and VOC2007. Experimental results demonstrate the MFB component efficiently fuses image representations and label co-occurrence embeddings and thus greatly improves the convergence efficiency of the model. In addition, the performance of image recognition has also been promoted compared with the state-of-the-art methods.

Configuration tuning is vital to optimize the performance of database management system (DBMS). It becomes more tedious and urgent for cloud databases (CDB) due to the diverse database instances and query workloads, which make the database administrator (DBA) incompetent. Although there are some studies on automatic DBMS configuration tuning, they have several limitations. Firstly, they adopt a pipelined learning model but cannot optimize the overall performance in an end-to-end manner. Secondly, they rely on large-scale high-quality training samples which are hard to obtain. Thirdly, there are a large number of knobs that are in continuous space and have unseen dependencies, and they cannot recommend reasonable configurations in such high-dimensional continuous space. Lastly, in cloud environment, they can hardly cope with the changes of hardware configurations and workloads, and have poor adaptability. To address these challenges, we design an end-to-end automatic CDB tuning system, CDBTune, using deep reinforcement learning (RL). CDBTune utilizes the deep deterministic policy gradient method to find the optimal configurations in high-dimensional continuous space. CDBTune adopts a try-and-error strategy to learn knob settings with a limited number of samples to accomplish the initial training, which alleviates the difficulty of collecting massive high-quality samples. CDBTune adopts the reward-feedback mechanism in RL instead of traditional regression, which enables end-to-end learning and accelerates the convergence speed of our model and improves efficiency of online tuning. We conducted extensive experiments under 6 different workloads on real cloud databases to demonstrate the superiority of CDBTune. Experimental results showed that CDBTune had a good adaptability and significantly outperformed the state-of-the-art tuning tools and DBA experts.

Many online social networks can be described by signed networks, where positive links signify friendships, trust and like; while negative links indicate enmity, distrust and dislike. Predicting the sign of the links in these networks has attracted a great deal of attentions in the areas of friendship recommendation and trust relationship prediction. Existing methods for sign prediction tend to rely on path-based features which are somehow limited to the sparsity problem of the network. In order to solve this issue, in this paper, we introduce a novel sign prediction model by exploiting cluster-based meta paths, which can take advantage of both local and global information of the input networks. First, cluster-based meta paths based features are constructed by incorporating the newly generated clusters through hierarchically clustering the input networks. Then, the logistic regression classifier is employed to train the model and predict the hidden signs of the links. Extensive experiments on Epinions and Slashdot datasets demonstrate the efficiency of our proposed method in terms of Accuracy and Coverage.