Modern analytics applications use a diverse mix of libraries and functions. Unfortunately, there is no optimization across these libraries, resulting in performance penalties as high as an order of magnitude in many applications. To address this problem, we proposed Weld, a common runtime for existing data analytics libraries that performs key physical optimizations such as pipelining under existing, imperative library APIs. In this work, we further develop the Weld vision by designing an automatic adaptive optimizer for Weld applications, and evaluating its impact on realistic data science workloads. Our optimizer eliminates multiple forms of overhead that arise when composing imperative libraries like Pandas and NumPy, and uses lightweight measurements to make data-dependent decisions at runtime in ad-hoc workloads where no statistics are available, with sub-second overhead. We also evaluate which optimizations have the largest impact in practice and whether Weld can be integrated into libraries incrementally. Our results are promising: using our optimizer, Weld accelerates data science workloads by up to 23× on one thread and 80× on eight threads, and its adaptive optimizations provide up to a 3.75× speedup over rule-based optimization. Moreover, Weld provides benefits if even just 4-5 operators in a library are ported to use it. Our results show that common runtime designs like Weld may be a viable approach to accelerate analytics.
A common operation in many data analytics workloads is to find the top-k items, i.e., the largest or smallest operations according to some sort order (implemented via LIMIT or ORDER BY expressions in SQL). A naive implementation of top-k is to sort all of the items and then return the first k, but this does much more work than needed. Although efficient implementations for top-k have been explored on traditional multi-core processors, there has been no prior systematic study of top-k implementations on GPUs, despite open requests for such implementations in GPU-based frameworks like TensorFlow 1 and ArrayFire 2 . In this work, we present several top-k algorithms for GPUs, including a new algorithm based on bitonic sort called bitonic top-k. The bitonic top-k algorithm is up to a factor of 15x faster than sort and 4x faster than a variety of other possible implementations for values of k up to 256. We also develop a cost model to predict the performance of several of our algorithms, and show that it accurately predicts actual performance on modern GPUs.
There has been significant amount of excitement and recent work on GPU-based database systems. Previous work has claimed that these systems can perform orders of magnitude better than CPU-based database systems on analytical workloads such as those found in decision support and business intelligence applications. A hardware expert would view these claims with suspicion. Given the general notion that database operators are memory-bandwidth bound, one would expect the maximum gain to be roughly equal to the ratio of the memory bandwidth of GPU to that of CPU. In this paper, we adopt a model-based approach to understand when and why the performance gains of running queries on GPUs vs on CPUs vary from the bandwidth ratio (which is roughly 16× on modern hardware). We propose Crystal, a library of parallel routines that can be combined together to run full SQL queries on a GPU with minimal materialization overhead. We implement individual query operators to show that while the speedups for selection, projection, and sorts are near the bandwidth ratio, joins achieve less speedup due to differences in hardware capabilities. Interestingly, we show on a popular analytical workload that full query performance gain from running on GPU exceeds the bandwidth ratio despite individual operators having speedup less than bandwidth ratio, as a result of limitations of vectorizing chained operators on CPUs, resulting in a 25× speedup for GPUs over CPUs on the benchmark.
Data partitioning is crucial to improving query performance and several workload-based partitioning techniques have been proposed in database literature. However, many modern analytic applications involve ad-hoc or exploratory analysis where users do not have a representative query workload a priori. Static workload-based data partitioning techniques are therefore not suitable for such settings. In this paper, we propose Amoeba, a distributed storage system that uses adaptive multi-attribute data partitioning to efficiently support ad-hoc as well as recurring queries. Amoeba requires zero set-up and tuning effort, allowing analysts to get the benefits of partitioning without requiring an upfront query workload. The key idea is to build and maintain a partitioning tree on top of the dataset. The partitioning tree allows us to answer queries with predicates by reading a subset of the data. The initial partitioning tree is created without requiring an upfront query workload and Amoeba adapts it over time by incrementally modifying subtrees based on user queries using repartitioning. A prototype of Amoeba running on top of Apache Spark improves query performance by up to 7x over full scans and up to 2x over range-based partitioning techniques on TPC-H as well as a real-world workload.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.