SIMD Vectorized Hashing for Grouped Aggregation

Gurumurthy, Bala; Broneske, David; Pinnecke, Marcus; Campero, Gabriel; Saake, Gunter

doi:10.1007/978-3-319-98398-1_8

Cited by 7 publications

(3 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They naturally extend and build upon each other, increasingly adjusted to vectorization. While vectorization of other hashing schemes such as Cuckoo and Hopscotch hashing has been considered [31,71], these works show that the benefit of vectorization is low and they fall into a separate family of hashing schemes.…”

Section: Vectorized Hashing Schemesmentioning

confidence: 99%

Analyzing Vectorized Hash Tables across CPU Architectures

Böther,

Benson,

Klimovic

et al. 2023

Proc. VLDB Endow.

View full text Add to dashboard Cite

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.

show abstract

Section: Vectorized Hashing Schemesmentioning

confidence: 99%

Analyzing Vectorized Hash Tables across CPU Architectures

Böther,

Benson,

Klimovic

et al. 2023

Proc. VLDB Endow.

View full text Add to dashboard Cite

show abstract

“…This is an extension of our previous work (Gurumurthy et al, 2023), where we include the concept and evaluation by investigating different caching techniques. Hence, we also contribute the investigation of:…”

Section: Introductionmentioning

confidence: 98%

Exploiting Shared Sub-Expression and Materialized View Reuse for Multi-Query Optimization

Gurumurthy,

Bidarkar,

Broneske

et al. 2024

Inf Syst Front

View full text Add to dashboard Cite

Querying in isolation lacks the potential of reusing intermediate results, which ends up wasting computational resources. Multi-Query Optimization (MQO) addresses this challenge by devising a shared execution strategy across queries, with two generally used strategies: batched or cached. These strategies are shown to improve performance, but hardly any study explores the combination of both. In this work we explore such a hybrid MQO, combining batching (Shared Sub-Expression) and caching (Materialized View Reuse) techniques. Our hybrid-MQO system merges batched query results as well as caches the intermediate results, thereby any new query is given a path within the previous plan as well as reusing the results. Since caching is a key component for improving performance, we measure the impact of common caching techniques such as FIFO, LRU, MRU and LFU. Our results show LRU to be the optimal for our usecase, which we use in our subsequent evaluations. To study the influence of batching, we vary the factor - - which represents the similarity of the results within a query batch. Similarly, we vary the cache sizes to study the influence of caching. Moreover, we also study the role of different database operators in the performance of our hybrid system. The results suggest that, depending on the individual operators, our hybrid method gains a speed-up between 4x to a slowdown of 2x from using MQO techniques in isolation. Furthermore, our results show that workloads with a generously sized cache that contain similar queries benefit from using our hybrid method, with an observed speed-up of 2x over sequential execution in the best case.

show abstract

“…This may incur a major cost factor [5][6][7]. Secondly, computing the grouping and aggregate is highly compute intensive [8][9][10], and thus a perfect use case for parallelization.…”

Section: Introductionmentioning

confidence: 99%

Novel insights on atomic synchronization for sort-based group-by on GPUs

Gurumurthy

Broneske

Schäler

et al. 2023

Distrib Parallel Databases

Self Cite

View full text Add to dashboard Cite

Using heterogeneous processing devices, like GPUs, to accelerate relational database operations is a well-known strategy. In this context, the operation is highly interesting for two reasons. Firstly, it incurs large processing costs. Secondly, its results (i.e., aggregates) are usually small, reducing data movement costs whose compensation is a major challenge for heterogeneous computing. Generally, for computation on GPUs, one relies either on sorting or hashing. Today, empirical results suggest that hash-based approaches are superior. However, by concept, hashing induces an unpredictable memory access pattern conflicting with the architecture of GPUs. This motivates studying why current sort-based approaches are generally inferior. Our results indicate that current sorting solutions cannot exploit the full parallel power of modern GPUs. Experimentally, we show that the issue arises from the need to synchronize parallel threads that access the shared memory location containing the aggregates via . Our quantification of the optimal performance motivates us to investigate how to minimize the overhead of atomics. This results in different variants using atomics, where the best variants almost mitigate the atomics overhead entirely. The results of a large-scale evaluation reveal that our approach achieves a 3x speed-up over existing sort-based approaches and up to 2x speed-up over hash-based approaches.

show abstract

SIMD Vectorized Hashing for Grouped Aggregation

Cited by 7 publications

References 12 publications

Analyzing Vectorized Hash Tables across CPU Architectures

Analyzing Vectorized Hash Tables across CPU Architectures

Exploiting Shared Sub-Expression and Materialized View Reuse for Multi-Query Optimization

Novel insights on atomic synchronization for sort-based group-by on GPUs

Contact Info

Product

Resources

About