Scanning HTML at Tens of Gigabytes per Second on ARM Processors

2025-03-03

AI Summary

To address performance bottlenecks in HTML parsing within Web browsers on ARM platforms, this paper proposes an efficient HTML scanning method specifically optimized for the ARM NEON instruction set. Unlike conventional vectorized classification approaches, our method introduces a novel SIMD acceleration mechanism that jointly exploits vectorized character classification and fine-grained parallel byte processing—while strictly preserving HTML semantic correctness. This design enables substantial throughput improvement without compromising parser accuracy or compliance. Experimental evaluation on state-of-the-art ARM processors achieves a peak HTML scanning throughput of 38 GB/s, representing a 20× speedup over the best prior solution. To the best of our knowledge, this work presents the first ultra-high-speed HTML scanner explicitly engineered for ARM architectures, establishing a new foundation for high-throughput, low-latency Web content parsing in modern ARM-based client and edge environments.

Technology Category

• Machine Learning: Hardware-aware ML
• Search and Optimization: Distributed Search
• Natural Language Processing: Syntax — Tagging, Chunking & Parsing

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization
• Web Mining and Content Analysis: Web measurements

Abstract

Modern processors have instructions to process 16 bytes or more at once. These instructions are called SIMD, for single instruction, multiple data. Recent advances have leveraged SIMD instructions to accelerate parsing of common Internet formats such as JSON and base64. During HTML parsing, they quickly identify specific characters with a strategy called vectorized classification. We review their techniques and compare them with a faster alternative. We measure a 20-fold performance improvement in HTML scanning compared to traditional methods on recent ARM processors. Our findings highlight the potential of SIMD-based algorithms for optimizing Web browser performance.

Problem

• Accelerate HTML parsing using SIMD instructions
• Compare vectorized classification with traditional methods
• Improve Web browser performance on ARM processors

Innovation

• Utilizes SIMD instructions for HTML parsing
• Implements vectorized classification for character identification
• Achieves 20x speedup on ARM processors