Sideways information passing is a well-known technique for mitigating the impact of large build sides in a database query plan. As currently implemented in production systems, sideways information passing enables only a uni-directional information flow, as opposed to instance-optimal algorithms, such as Yannakakis'. On the other hand, the latter require an additional pass over the input, which hinders adoption in production systems. In this paper, we make a step towards enabling single-pass bidirectional information passing during query execution. We achieve this by statically analyzing between which tables the information flow is blocked and by leveraging precomputed join-induced fingerprint columns on FK-tables. On the JOB benchmark, Parachute improves DuckDB v1.2's end-to-end execution time without and with semi-join filtering by 1.54x and 1.24x, respectively, when allowed to use 15% extra space.

We revisit the join ordering problem in query optimization. The standard exact algorithm, DPccp, has a worst-case running time of O(3 n ). This is prohibitively expensive for large queries, which are not that uncommon anymore. We develop a new algorithmic framework based on subset convolution. DPconv achieves a super-polynomial speedup over DPccp, breaking the O(3 n ) time-barrier for the first time. We show that the framework instantiation for the C max cost function is up to 30x faster than DPccp for large clique queries.

Database research and development is heavily influenced by benchmarks, such as the industry-standard TPC-H and TPC-DS for analytical systems. However, these twenty-year-old benchmarks neither capture how databases are deployed nor what workloads modern cloud data warehouse systems face these days. In this paper, we summarize well-known, confirm suspected, and unearth novel discrepancies between TPC-H/DS and actual workloads using empirical data. We base our analysis on telemetrics from Amazon Redshift - one of the largest cloud data warehouse deployments. Among others, we show how write-heavy data pipelines are prominent, workloads vary over time (in both load and type), queries are repetitive, and how most properties of queries or workloads experience very long tailed distributions. We conclude that data warehouse benchmarks, just like database systems, need to become more holistic and stop focusing solely on query engine performance. Finally, we publish a dataset containing query statistics of 200 randomly selected Redshift serverless and provisioned instances (each) over a three-month period, as a basis for building more realistic benchmarks.

Schema discovery and data loading is a crucial step in any data analysis pipeline. While this used to be a rare task, in the highly dynamic field of machine learning and modern business intelligence on top of data lakes, today it has become a frequent, but often underestimated, activity. Existing tools often focus on single files, presume prior knowledge of the data on the user's side or a significant amount of manual labor. In this paper, we improve the process of mapping a "chaotic" set of files to an initial database schema that can then be iteratively refined and loaded. The idea is to take the previously tedious parts of this process and automate them through the use of Large Language Models (LLMs) while leaving already well-understood problems such as constraint discovery to existing algorithms. We thus carefully orchestrate the use of LLMs for the "soft" problems and traditional algorithms for the "hard" problems. This creates a more seamless schema discovery and data loading experience that minimizes the time to first insight for users. We show this vision on modern schema discovery and data loading in our web-based prototype called DataLoom that serves as our demonstration.

Query driven cardinality estimation models learn from a historical log of queries. They are lightweight, having low storage requirements, fast inference and training, and are easily adaptable for any kind of query. Unfortunately, such models can suffer unpredictably bad performance under workload drift, i.e., if the query pattern or data changes. This makes them unreliable and hard to deploy. We analyze the reasons why models become unpredictable due to workload drift, and introduce modifications to the query representation and neural network training techniques to make query-driven models robust to the effects of workload drift. First, we emulate workload drift in queries involving some unseen tables or columns by randomly masking out some table or column features during training. This forces the model to make predictions with missing query information, relying more on robust features based on up-to-date DBMS statistics that are useful even when query or data drift happens. Second, we introduce join bitmaps, which extends sampling-based features to be consistent across joins using ideas from sideways information passing. Finally, we show how both of these ideas can be adapted to handle data updates. We show significantly greater generalization than past works across different workloads and databases. For instance, a model trained with our techniques on a simple workload (JOBLight-train), with 40 k synthetically generated queries of at most 3 tables each, is able to generalize to the much more complex Join Order Benchmark, which include queries with up to 16 tables, and improve query runtimes by 2× over PostgreSQL. We show similar robustness results with data updates, and across other workloads. We discuss the situations where we expect, and see, improvements, as well as more challenging workload drift scenarios where these techniques do not improve much over PostgreSQL. However, even in the most challenging scenarios, our models never perform worse than PostgreSQL, while standard query driven models can get much worse than PostgreSQL.

While index structures are crucial components in high-performance query processing systems, they occupy a large fraction of the available memory. Recently-proposed compact indexes reduce this space overhead and thus speed up queries by allowing the database to keep larger working sets in memory. These compact indexes, however, are slower than performance-optimized in-memory indexes because they adopt encodings that trade performance for memory efficiency. Applying different encodings within a single index might allow optimizing both dimensions at the same time - however, it is not clear which encodings should be applied to which index parts at build-time.

Modern data systems are typically both complex and general-purpose. They are complex because of the numerous internal knobs and parameters that users need to manually tune in order to achieve good performance; they are general-purpose because they are designed to handle diverse use cases, and therefore often do not achieve the best possible performance for any specific use case. A recent trend aims to tackle these pitfalls: instance-optimized systems are designed to automatically self-adjust in order to achieve the best performance for a specific use case, i.e., a dataset and query workload. Thus far, the research community has focused on creating instance-optimized database components, such as learned indexes and learned cardinality estimators, which are evaluated in isolation. However, to the best of our knowledge, there is no complete data system built with instance-optimization as a foundational design principle. In this paper, we present a progress report on SageDB, our effort towards building the first instance-optimized data system. SageDB synthesizes various instance-optimization techniques to automatically specialize for a given use case, while simultaneously exposing a simple user interface that places minimal technical burden on the user. Our prototype outperforms a commercial cloud-based analytics system by up to 3X on end-to-end query workloads and up to 250X on individual queries. SageDB is an ongoing research effort, and we highlight our lessons learned and key directions for future work.

Hashing is a fundamental operation in database management, playing a key role in the implementation of numerous core database data structures and algorithms. Traditional hash functions aim to mimic a function that maps a key to a random value, which can result in collisions, where multiple keys are mapped to the same value. There are many well-known schemes like chaining, probing, and cuckoo hashing to handle collisions. In this work, we aim to study if using learned models instead of traditional hash functions can reduce collisions and whether such a reduction translates to improved performance, particularly for indexing and joins. We show that learned models reduce collisions in some cases, which depend on how the data is distributed. To evaluate the effectiveness of learned models as hash function, we test them with bucket chaining, linear probing, and cuckoo hash tables. We find that learned models can (1) yield a 1.4x lower probe latency, and (2) reduce the non-partitioned hash join runtime with 28% over the next best baseline for certain datasets. On the other hand, if the data distribution is not suitable, we either do not see gains or see worse performance. In summary, we find that learned models can indeed outperform hash functions, but only for certain data distributions.

Many modern key-value stores, such as RocksDB, rely on log-structured merge trees (LSMs). Originally designed for spinning disks, LSMs optimize for write performance by only making sequential writes. But this optimization comes at the cost of reads: LSMs must rely on expensive compaction jobs and Bloom filters---all to maintain reasonable read performance. For NVMe SSDs, we argue that trading off read performance for write performance is no longer always needed. With enough parallelism, NVMe SSDs have comparable random and sequential access performance. This change makes update-in-place designs, which traditionally provide excellent read performance, a viable alternative to LSMs. In this paper, we close the gap between log-structured and update-in-place designs on modern SSDs with the help of new components that take advantage of data and workload patterns. Specifically, we explore three key ideas: (A) record caching for efficient point operations, (B) page grouping for high-performance range scans, and (C) insert forecasting to reduce the reorganization costs of accommodating new records. We evaluate these ideas by implementing them in a prototype update-in-place key-value store called TreeLine. On YCSB, we find that TreeLine outperforms RocksDB and LeanStore by 2.20× and 2.07× respectively on average across the point workloads, and by up to 10.95× and 7.52× overall.

Abstract Many applications today like Uber, Yelp, Tinder, etc. rely on spatial data or locations from its users. These applications and services either build their own spatial data management systems or rely on existing solutions. JTS Topology Suite (JTS), its C++ port GEOS, Google S2, ESRI Geometry API, and Java Spatial Index (JSI) are some of the spatial processing libraries that these systems build upon. These applications and services depend on indexing capabilities available in these libraries for high-performance spatial query processing. In this work, we compare these libraries qualitatively and quantitatively based on four different spatial queries using two real world datasets. We also compare these libraries with an open-source implementation of the Vantage Point Tree—an index structure that has been well studied in image retrieval and nearest-neighbor search algorithms for high-dimensional data. We found that Vantage Point Trees are very competitive and even outperform the aforementioned libraries in two queries.

Recently there has been significant interest in using machine learning to improve the accuracy of cardinality estimation. This work has focused on improving average estimation error, but not all estimates matter equally for downstream tasks like query optimization. Since learned models inevitably make mistakes, the goal should be to improve the estimates that make the biggest difference to an optimizer. We introduce a new loss function, Flow-Loss, for learning cardinality estimation models. Flow-Loss approximates the optimizer's cost model and search algorithm with analytical functions, which it uses to optimize explicitly for better query plans. At the heart of Flow-Loss is a reduction of query optimization to a flow routing problem on a certain "plan graph", in which different paths correspond to different query plans. To evaluate our approach, we introduce the Cardinality Estimation Benchmark (CEB) which contains the ground truth cardinalities for sub-plans of over 16 K queries from 21 templates with up to 15 joins. We show that across different architectures and databases, a model trained with Flow-Loss improves the plan costs and query runtimes despite having worse estimation accuracy than a model trained with Q-Error. When the test set queries closely match the training queries, models trained with both loss functions perform well. However, the Q-Error-trained model degrades significantly when evaluated on slightly different queries (e.g., similar but unseen query templates), while the Flow-Loss-trained model generalizes better to such situations, achieving 4 -- 8× better 99th percentile runtimes on unseen templates with the same model architecture and training data.

Connected mobility applications rely heavily on geospatial joins that associate point data, such as locations of Uber cars, to static polygonal regions, such as city neighborhoods. These joins typically involve expensive geometric computations, which makes it hard to provide an interactive user experience. In this paper, we propose an adaptive polygon index that leverages true hit fltering to avoid expensive geometric computations in most cases. In particular, our approach closely approximates polygons by combining quadtrees with true hit filtering, and stores these approximations in a query-effcient radix tree. Based on this index, we introduce two geospatial join algorithms: an approximate one that guarantees a user-defined precision, and an exact one that adapts to the expected point distribution. In summary, our technique outperforms existing CPU-based joins by up to two orders of magnitude and is competitive with state-of-the-art GPU implementations.

While hardware and software improvements greatly accelerated modern database systems' internal operations, the decades-old stream-based Socket API for external communication is still unchanged. We show experimentally, that for modern high-performance systems networking has become a performance bottleneck. Therefore, we argue that the communication stack needs to be redesigned to fully exploit modern hardware - as has already happened to most other database system components.We propose L5, a high-performance communication layer for database systems. L5 rethinks the flow of data in and out of the database system and is based on direct memory access techniques for intra-datacenter (RDMA) and intra-machine communication (Shared Memory). With L5, we provide a building block to accelerate ODBC-like interfaces with a unified and message-based communication framework. Our results show that using interconnects like RDMA (InfiniBand), RoCE (Ethernet), and Shared Memory (IPC), L5 can largely eliminate the network bottleneck for database systems.

In modern data warehousing, data skipping is essential for high query performance. While index structures such as B-trees or hash tables allow for precise pruning, their large storage requirements make them impractical for indexing secondary columns. Therefore, many systems rely on approximate indexes such as min/max sketches (ZoneMaps) or Bloom filters for cost-effective data pruning. For example, Google PowerDrill skips more than 90% of data on average using such indexes. In this paper, we introduce Cuckoo Index (CI), an approximate secondary index structure that represents the many-to-many relationship between keys and data partitions in a highly space-efficient way. At its core, CI associates variable-sized fingerprints in a Cuckoo filter with compressed bitmaps indicating qualifying partitions. With our approach, we target equality predicates in a read-only (immutable) setting and optimize for space efficiency under the premise of practical build and lookup performance. In contrast to per-partition (Bloom) filters, CI produces correct results for lookups with keys that occur in the data. CI allows to control the ratio of false positive partitions for lookups with non-occurring keys. Our experiments with real-world and synthetic data show that CI consumes significantly less space than per-partition filters for the same pruning power for low-to-medium cardinality columns. For high cardinality columns, CI is on par with its baselines.

Recent advancements in learned index structures propose replacing existing index structures, like B-Trees, with approximate learned models. In this work, we present a unified benchmark that compares well-tuned implementations of three learned index structures against several state-of-the-art "traditional" baselines. Using four real-world datasets, we demonstrate that learned index structures can indeed outperform non-learned indexes in read-only in-memory workloads over a dense array. We investigate the impact of caching, pipelining, dataset size, and key size. We study the performance profile of learned index structures, and build an explanation for why learned models achieve such good performance. Finally, we investigate other important properties of learned index structures, such as their performance in multi-threaded systems and their build times.

Increasing single instruction multiple data (SIMD) capabilities in modern hardware allows for the compilation of data-parallel query pipelines. This means GPU-alike challenges arise: control flow divergence causes the underutilization of vector-processing units. In this paper, we present efficient algorithms for the AVX-512 architecture to address this issue. These algorithms allow for the fine-grained assignment of new tuples to idle SIMD lanes. Furthermore, we present strategies for their integration with compiled query pipelines so that tuples are never evicted from registers. We evaluate our approach with three query types: (i) a table scan query based on TPC-H Query 1, that performs up to 34% faster when addressing underutilization, (ii) a hashjoin query, where we observe up to 25% higher performance, and (iii) an approximate geospatial join query, which shows performance improvements of up to 30%.

The amount of available geospatial data grows at an ever faster pace. This leads to a constantly increasing demand for processing power and storage in order to provide data analysis in a timely manner. At the same time, a lot of geospatial processing is visual and exploratory in nature, thus having bounded precision requirements. We present DeepSPACE, a deep learning-based approximate geospatial query processing engine which combines modest hardware requirements with the ability to answer flexible aggregation queries while keeping the required state to a few hundred KiBs.

We introduce Deep Sketches, which are compact models of databases that allow us to estimate the result sizes of SQL queries. Deep Sketches are powered by a new deep learning approach to cardinality estimation that can capture correlations between columns, even across tables. Our demonstration allows users to define such sketches on the TPC-H and IMDb datasets, monitor the training process, and run ad-hoc queries against trained sketches. We also estimate query cardinalities with HyPer and PostgreSQL to visualize the gains over traditional cardinality estimators.

Today’s streaming applications demand increasingly high event throughput rates and are often subject to strict latency constraints. To allow for more complex workloads, such as window-based aggregations, streaming systems need to support stateful event processing. This introduces new challenges for streaming engines as the state needs to be maintained in a consistent and durable manner and simultaneously accessed by complex queries for real-time analytics. Modern streaming systems, such as Apache Flink, do not allow for efficiently exposing the state to analytical queries. Thus, data engineers are forced to keep the state in external data stores, which significantly increases the latencies until events become visible to analytical queries. Proprietary solutions have been created to meet data freshness constraints. These solutions are expensive, error-prone, and difficult to maintain. Main-memory database systems, such as HyPer, achieve extremely low query response times while maintaining high update rates, which makes them well-suited for analytical streaming workloads. In this article, we explore extensions to database systems to match the performance and usability of streaming systems.

Increasing single instruction multiple data (SIMD) capabilities in modern hardware allows for compiling efficient data-parallel query pipelines. This means GPU-alike challenges arise: control flow divergence causes underutilization of vector-processing units. In this paper, we present efficient algorithms for the AVX-512 architecture to address this issue. These algorithms allow for fine-grained assignment of new tuples to idle SIMD lanes. Furthermore, we present strategies for their integration with compiled query pipelines without introducing inefficient memory materializations. We evaluate our approach with a high-performance geospatial join query, which shows performance improvements of up to 35%.

Geospatial joins are a core building block of connected mobility applications. An especially challenging problem are joins between streaming points and static polygons. Since points are not known beforehand, they cannot be indexed. Nevertheless, points need to be mapped to polygons with low latencies to enable real-time feedback. We present an approximate geospatial join that guarantees a user-defined precision. Our technique uses a quadtree-based hierarchical grid to approximate polygons and stores these approximations in a specialized radix tree. Our approach can perform up to several orders of magnitude faster than existing techniques while providing sufficiently precise results for many applications.

Spatial data is pervasive. Large amount of spatial data is produced every day from GPS-enabled devices such as cell phones, cars, sensors, and various consumer based applications such as Uber, location-tagged posts in Facebook, In-stagram, Snapchat, etc. This growth in spatial data coupled with the fact that spatial queries, analytical or transactional, can be computationally extensive has attracted enormous interest from the research community to develop systems that can efficiently process and analyze this data. In recent years a lot of spatial analytics systems have emerged. Existing work compares either limited features of these systems or the studies are outdated since new systems have emerged. In this work, we first explore the available modern spatial processing systems and then thoroughly compare them based on features and queries they support, using real-world datasets.

In the past few years, massive amounts of location-based data has been captured. Numerous datasets containing user location information are readily available to the public. Analyzing such datasets can lead to fascinating insights into the mobility patterns and behaviors of users. Moreover, in recent times a number of geospatial data-driven companies like Uber, Lyft, and Foursquare have emerged. Real-time analysis of geospatial data is essential and enables an emerging class of applications. Database support for geospatial operations is turning into a necessity instead of a distinct feature provided by only a few databases. Even though a lot of database systems provide geospatial support nowadays, queries often do not consider the most current database state. Geospatial queries are inherently slow given the fact that some of these queries require a couple of geometric computations. Disk-based database systems that do support geospatial datatypes and queries, provide rich features and functions, but they fall behind when performance is considered: specifically if real-time analysis of the latest transactional state is a requirement. In this demonstration, we present HyPerSpace, an extension to the high-performance main-memory database system HyPer developed at the Technical University of Munich, capable of processing geospatial queries with sub-second latencies.