The query execution methods in database systems are continuously evolving to leverage emerging hardware capabilities aimed at enhancing performance, particularly for analytical workloads.

Over the past decades, the Single Instruction Multiple Data (SIMD) paradigm has emerged as a leading strategy to boost single-query performance through in-core parallelization.

Current advancements in SIMD capabilities within modern processors are characterized by increased data-level parallelism and enhanced functional abilities.

This tutorial provides a contemporary overview of integrating SIMD features into data-intensive analytical processing.

The structure of the tutorial is divided into three key parts.

The first section establishes foundational concepts by elucidating SIMD—its origins, intended purposes, and traditional applications in accelerating query processing.

In the second section, we present a framework for writing hardware-agnostic explicit SIMD code utilizing the TSL, a SIMD-hardware abstraction library.

The final part elaborates on extending the TSL to accommodate novel architectures, exploring SIMD concepts such as vector-length agnostic programming, and extending support to non-traditional hardware platforms including FPGAs.

Over the past decade, storage technology has evolved significantly.

Traditional rotational magnetic media, which offered hundreds of IOPS and latencies measured in milliseconds, have been largely replaced by SSDs.

Modern SSDs deliver over 10 million IOPS with latencies in the low double-digit microsecond range.

This shift has disrupted the assumptions underlying the design of traditional I/O stacks, making them less suitable for modern hardware.

To address these challenges, new APIs like NVMe (at the protocol level) and io_uring (at the operating system level) have been developed.

These APIs aim to eliminate performance bottlenecks that have become apparent as the hardware has advanced.

At the OS level, a key focus of these APIs is reducing the number of system calls, as excessive syscalls can severely impact performance on modern systems by disrupting processor caches.

Beyond this, many modern APIs share common design principles, whether they operate at the protocol, OS, or user-space library level.

They are typically asynchronous, completion-based, and optimized for high-concurrency I/O operations to fully leverage the capabilities of current hardware.

In our proposed tutorial, we will begin by introducing io_uring and explaining its basic functionality.

We will highlight its advantages compared to older I/O APIs and conclude with a discussion of the challenges faced by user-space applications seeking to improve performance with io_uring.

These include the increased complexity of implementation and scaling issues related to parallelism.