KScaNN: Scalable Approximate Nearest Neighbor Search on Kunpeng

TL;DR

KScaNN presents a hardware-aware, co-designed approach to scalable approximate nearest neighbor search on Kunpeng ARM systems. By integrating ML-driven adaptive search, a hybrid intra-cluster graph strategy, and highly optimized ARM SIMD kernels for PQ distance calculations, it closes the ARM-x86 performance gap and achieves up to $1.63\times$ speedup over leading x86 baselines. The work demonstrates that data-aware dimensionality reduction, per-query parameter tuning, and tight hardware specialization are crucial for high-throughput vector search on modern ARM CPUs. Its findings offer a practical blueprint for deploying leadership-class ANNS on ARM infrastructure and highlight the importance of co-design between algorithms and heterogeneous hardware.

Abstract

Approximate Nearest Neighbor Search (ANNS) is a cornerstone algorithm for information retrieval, recommendation systems, and machine learning applications. While x86-based architectures have historically dominated this domain, the increasing adoption of ARM-based servers in industry presents a critical need for ANNS solutions optimized on ARM architectures. A naive port of existing x86 ANNS algorithms to ARM platforms results in a substantial performance deficit, failing to leverage the unique capabilities of the underlying hardware. To address this challenge, we introduce KScaNN, a novel ANNS algorithm co-designed for the Kunpeng 920 ARM architecture. KScaNN embodies a holistic approach that synergizes sophisticated, data aware algorithmic refinements with carefully-designed hardware specific optimizations. Its core contributions include: 1) novel algorithmic techniques, including a hybrid intra-cluster search strategy and an improved PQ residual calculation method, which optimize the search process at a higher level; 2) an ML-driven adaptive search module that provides adaptive, per-query tuning of search parameters, eliminating the inefficiencies of static configurations; and 3) highly-optimized SIMD kernels for ARM that maximize hardware utilization for the critical distance computation workloads. The experimental results demonstrate that KScaNN not only closes the performance gap but establishes a new standard, achieving up to a 1.63x speedup over the fastest x86-based solution. This work provides a definitive blueprint for achieving leadership-class performance for vector search on modern ARM architectures and underscores

Figures (13)

• Figure 1: The pipeline of KScaNN algorithm
• Figure 2: Proportion of queries with 10 NN found for recall 0.99 for SIFT1M dataset.
• Figure 3: Distribution of the top-100 nearest neighbors across the most proximate leaf clusters for a sample query on the GLOVE-100 dataset. Each column represents a cluster, revealing that most probed clusters are non-contributory.
• Figure 4: Illustration of cluster pruning scenarios. FP: A cluster whose centroid is close to the query while its points stretch in an orthogonal direction, which is unlikely to contain nearest neighbors. TP: A cluster with a relatively distant centroid that contains outlier points which are true nearest neighbors to the query.
• Figure 5: Sample images from the FASHION-MNIST dataset, characterized by large, uninformative background regions that correspond to low-variance dimensions.