Search for: All records

Query optimization is the process of finding an efficient query execution plan for a given SQL query. The runtime difference between a good and a bad plan can be tremendous. For example, in the case of TPC-H query 5, a query with 5 joins, the difference between the best and the worst plan is more than 10,000×. Therefore, it is vital to avoid bad plans. The dominating factor which differentiates a good from a bad plan is their join order and whether this join order avoids large intermediate results. 

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 40ksynthetically 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. 

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. 

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. 

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.