1.
Mahling, F., Rößler, P., Bodner, T., Rabl, T.: BabelMR: A Polyglot Framework for Serverless MapReduce. Workshop on Serverless Data Analytics. (2023).
The MapReduce programming model and its open-source implementation Hadoop have democratized large-scale data processing by providing ease-of-use and scalability. Subsequently, systems such as Spark have dramatically improved efficiency. However, for a large number of users and applications, using these frameworks remains challenging, because they typically restrict them to specific programming languages or require cluster management expertise. In this paper, we present BabelMR, a data processing framework that provides the MapReduce programming model to arbitrary containerized applications to be executed on serverless cloud infrastructure. Users provide application logic in Map and Reduce functions that read and write their inputs and outputs to the ephemeral filesystem of a serverless function container. BabelMR orchestrates the data-parallel programs across stages of concurrent cloud function executions and efficiently integrates with serverless storage systems and columnar storage formats. Our evaluation shows that BabelMR reduces the entry hurdle to analyzing data in a distributed serverless environment in terms of development effort. BabelMR’s I/O and data shuffle building blocks outperform handwritten Python and C# code, and BabelMR is competitive with state-of- the-art serverless MapReduce systems.
2.
Benson, L., Ebeling, R., Rabl, T.: Evaluating SIMD Compiler-Intrinsics for Database Systems. 14th International Workshop on Accelerating Analytics and Data Management Systems Using Modern Processors and Storage Architectures. (2023).
Modern query engines often use SIMD instructions to speed up query performance. As these instructions are heavily CPU-specific, developers must write multiple variants of the same code to support multiple target platforms such as AVX2, AVX512, and ARM NEON. This process leads to logical code duplication, which is cumbersome, hard to test, and hard to benchmark. In this paper, we make the case for writing less platform-specific SIMD code by leveraging the compiler’s own platform-independent SIMD vector abstraction. This allows developers to write a single code variant for all platforms as with a SIMD library, without the library’s redundant layers of abstraction. Clang and GCC implement the platforms’ SIMD intrinsics on top of their own abstraction, so code written in it is optimized for the underlying vector instructions by the compiler. We conduct four database operation microbenchmarks based on code in real systems on x86 and ARM and show that compiler-intrinsic variants achieve the same or even better performance than platform-intrinsics in most cases. In addition, we completely replace the SIMD library in the state-of-the-art query engine Velox with compiler-intrinsics. Our results show that query engines can achieve the same performance with platform-independent code while requiring significantly less SIMD code and fewer variants.
3.
Brücke, C., Härtling, P., Escobar Palacios, R.D., Patel, H., Rabl, T.: TPCx-AI - An Industry Standard Benchmark for Artificial Intelligence and Machine Learning Systems. Proceedings of the VLDB Endowment. 16, 3649–3661 (2023).
Artificial intelligence (AI) and machine learning (ML) techniques have existed for years, but new hardware trends and advances in model training and inference have radically improved their perfor- mance. With an ever increasing amount of algorithms, systems, and hardware solutions, it is challenging to identify good deployments even for experts. Researchers and industry experts have observed this challenge and have created several benchmark suites for AI and ML applications and systems. While they are helpful in comparing several aspects of AI applications, none of the existing benchmarks measures end-to-end performance of ML deployments. Many have been rigorously developed in collaboration between academia and industry, but no existing benchmark is standardized. In this paper, we introduce the TPC Express Benchmark for Arti- ficial Intelligence (TPCx-AI), the first industry standard benchmark for end-to-end machine learning deployments. TPCx-AI is the first AI benchmark that represents the pipelines typically found in com- mon ML and AI workloads. TPCx-AI provides a full software kit, which includes data generator, driver, and two full workload imple- mentations, one based on Python libraries and one based on Apache Spark. We describe the complete benchmark and show benchmark results for various scale factors. TPCx-AI’s core contributions are a novel unified data set covering structured and unstructured data; a fully scalable data generator that can generate realistic data from GB up to PB scale; and a diverse and representative workload using different data types and algorithms, covering a wide range of as- pects of real ML workloads such as data integration, data processing, training, and inference.
4.
Böther, M., Benson, L., Klimovic, A., Rabl, T.: Analyzing Vectorized Hash Tables Across CPU Architectures. Proceedings of the VLDB Endowment. 16, 2755–2768 (2023).
Data processing systems often leverage vector instructions to achieve higher performance.When applying vector instructions, an often overlooked data structure is the hash table, even though it is fundamental in data processing systems for operations such as indexing, aggregating, and joining. In this paper, we characterize and evaluate three fundamental vectorized hashing schemes, vectorized linear probing (VLP), vectorized fingerprinting (VFP), and bucket-based comparison (BBC). We implement these hashing schemes on the x86, ARM, and Power CPU architectures, as modern database systems must provide efficient implementations for multiple platforms due to the continuously increasing hardware heterogeneity. We present various implementation variants and platform-specific optimizations, which we evaluate for integer keys, string keys, large payloads, skewed distributions, and multiple threads. Our extensive evaluation and comparison to three scalar hashing schemes on four servers shows that BBC outperforms scalar linear probing by a factor of more than 2x, while also scaling well to high load factors. We find that vectorized hashing schemes come with caveats that need to be considered, such as the increased engineering overhead, differences between CPUs, and differences between vector ISAs, such as AVX and AVX-512, which impact performance. We conclude with key findings for vectorized hashing scheme implementations.
5.
Wang, Y., Benson, L., Rabl, T.: Desis: Efficient Window Aggregation in Decentralized Networks. 26th International Conference on Extending Database Technology (EDBT ’23) (2023).
Stream processing is widely applied in industry as well as in research to process unbounded data streams. In many use cases, specific data streams are processed by multiple continuous queries. Current systems group events of an unbounded data stream into bounded windows to produce results of individual queries in a timely fashion. For multiple concurrent queries, multiple concurrent and usually overlapping windows are generated. To reduce redundant computations and share partial results, state-of-the-art solutions divide windows into slices and then share the results of those slices. However, this is only applicable for queries with the same aggregation function and window measure, as in the case of overlaps for sliding windows. For multiple queries on the same stream with different aggregation functions and window measures, partial results cannot be shared. Furthermore, data streams are produced from devices that are distributed in large decentralized networks. Current systems cannot process queries on decentralized data streams efficiently. All queries in a decentralized network are either computed centrally or processed individually without exploiting partial results across queries. We present Desis, a stream processing system that can efficiently process multiple stream aggregation queries. We propose an aggregation engine that can share partial results between multiple queries with different window types, measures, and aggregation functions. In decentralized networks, Desis moves computation to data sources and shares overlapping computation as early as possible between queries. Desis outperforms existing solutions by orders of magnitude in throughput when processing multiple queries and can scale to millions of queries. In a decentralized setup, Desis can save up to 99% of network traffic and scale performance linearly.
6.
Strassenburg, N., Kupfer, D., Kowal, J., Rabl, T.: Efficient Multi-Model Management. 26th International Conference on Extending Database Technology (EDBT ’23). (2023).
Deep Learning models are deployed in an increasing number of industrial domains, such as retail and automotive applications. An instance of a model typically performs one specific task, which is why larger software systems use multiple models in parallel. Given that all models in production software have to be managed, this leads to the problem of managing sets of related models, i.e., multi-model management. Existing approaches perform poorly on this task because they are optimized for saving single large models but not for simultaneously saving a set of related models. In this paper, we explore the space of multi-model management by presenting three optimized approaches: (1) A baseline approach that saves full model representations and minimizes the amount of saved metadata. (2) An update approach that reduces the storage consumption compared to the baseline by saving parameter updates instead of full models. (3) A provenance approach that saves model provenance data instead of model parameters. We evaluate the approaches for the multi-model management use cases of managing car battery cell models and image classification models. Our results show that the baseline outperforms existing approaches for save and recover times by more than an order of magnitude and that more sophisticated approaches reduce the storage consumption by up to 99%.
7.
Ilic, I., Tolovski, I., Rabl, T.: RMG Sort: Radix-Partitioning-Based Multi-GPU Sorting. In: et al., B.K.-R. (ed.) Datenbanksysteme für Business, Technologie und Web (BTW 2023) (2023).
In recent years, graphics processing units (GPUs) emerged as database accelerators due to their massive parallelism and high-bandwidth memory. Sorting is a core database operation with many applications, such as output ordering, index creation, grouping, and sort-merge joins. Many single-GPU sorting algorithms have been shown to outperform highly parallel CPU algorithms. Today’s systems include multiple GPUs with direct high-bandwidth peer-to-peer (P2P) interconnects. However, previous multi-GPU sorting algorithms do not efficiently harness the P2P transfer capability of modern interconnects, such as NVLink and NVSwitch. In this paper, we propose RMG sort, a novel radix partitioning-based multi-GPU sorting algorithm. We present a most-significant-bit partitioning strategy that efficiently utilizes high-speed P2P interconnects while reducing inter-GPU communication. Independent of the number of GPUs, we exchange radix partitions between the GPUs in one all-to-all P2P key swap and achieve nearly-perfect load balancing. We evaluate RMG sort on two modern multi-GPU systems. Our experiments show that RMG sort scales well with the input size and the number of GPUs, outperforming a parallel CPU-based sort by up to 20×. Compared to two state-of-the-art, merge-based, multi-GPU sorting algorithms, we achieve speedups of up to 1.3× and 1.8× across both systems. Excluding the CPU-GPU data transfer times and on eight GPUs, RMG sort outperforms the two merge-based multi-GPU sorting algorithms up to 2.7× and 9.2×.
