Voodoo - a vector algebra for portable database performance on modern hardware

Abstract

In-memory databases require careful tuning and many engineering tricks to achieve good performance. Such database performance engineering is hard: a plethora of data and hardware-dependent optimization techniques form a design space that is difficult to navigate for a skilled engineer-even more so for a query compiler. To facilitate performanceoriented design exploration and query plan compilation, we present Voodoo, a declarative intermediate algebra that abstracts the detailed architectural properties of the hardware, such as multi-or many-core architectures, caches and SIMD registers, without losing the ability to generate highly tuned code. Because it consists of a collection of declarative, vector-oriented operations, Voodoo is easier to reason about and tune than low-level C and related hardware-focused extensions (Intrinsics, OpenCL, CUDA, etc.). This enables our Voodoo compiler to produce (OpenCL) code that rivals and even outperforms the fastest state-of-the-art in memory databases for both GPUs and CPUs. In addition, Voodoo makes it possible to express techniques as diverse as cacheconscious processing, predication and vectorization (again on both GPUs and CPUs) with just a few lines of code. Central to our approach is a novel idea we termed control vectors, which allows a code generating frontend to expose parallelism to the Voodoo compiler in a abstract manner, enabling portable performance across hardware platforms. We used Voodoo to build an alternative backend for Mon-etDB, a popular open-source in-memory database. Our backend allows MonetDB to perform at the same level as highly tuned in-memory databases, including HyPeR and Ocelot. We also demonstrate Voodoo's usefulness when investigating hardware conscious tuning techniques, assessing their performance on different queries, devices and data.

On optimizing operator fusion plans for large-scale machine learning in systemML

“…Similarly, the literature on optimizing fusion plans is very sparse. Frameworks such as OptiML [88], Emma [2], Kasen [94], Voodoo [73], SystemML-SPOOF [27], Weld [70], and TensorFlow XLA [1,33] all use fusion heuristics, which misses significant opportunities. Tupleware [22] combines heuristics and cost-based decisions for micro-optimizations such as predication and loop tiling.…”

Generating custom code for efficient query execution on heterogeneous processors

“…Pirk and others propose the Voodoo framework, which consists of an intermediate algebra representation based on vectors and a code generator for OpenCL [39]. Based on the algebra, Voodoo is capable of generating code for different processors, including CPUs and GPUs.…”

Designing an Adaptive VM That Combines Vectorized and JIT Execution on Heterogeneous Hardware

“…Unlike VOODOO [23], Weld [19] and MRQL [8] it includes control-flow operations. This allows the program to be implemented using specialized operations on chunked data (i.e.…”

“…Previous approaches, such as VOODOO [23], Weld [19] and MRQL [8], proposed domain-specific languages (DSLs) that abstract implementation and hardware-specific details away. However, these attempts impose two challenges: First, the simplifying assumptions these make on data storage to make data fit across all platforms make it hard to leverage existing data ordering/clustering and, second, there is no generic way to exploit specialized instructions as this requires matching a code pattern to the instruction's specific use-case.…”