Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HDDB: Efficient In-Storage SQL Database Search Using Hyperdimensional Computing on Ferroelectric NAND Flash

Quanling Zhao, Yanru Chen, Runyang Tian, Sumukh Pinge, Weihong Xu, Augusto Vega, Steven Holmes, Saransh Gupta, Tajana Rosing

TL;DR

HDDB addresses the energy- and bandwidth-intensive predicate evaluation in SQL analytics over large fact tables by co-designing Hyperdimensional Computing with FeNAND in-storage processing. It encodes SQL tables into hypervectors and implements in-storage predicate evaluation and decoding using HDC primitives, achieving noise robustness and high parallelism. The system introduces a columnar, in-storage mapping, a DBAM-based approximate matching accelerator, and specialized near-storage peripherals, delivering up to 80.6× latency reduction and 12,636× energy savings over CPU/GPU baselines while tolerating up to 10% random TLC cell corruption. This work demonstrates a practical memory-centric database processing substrate and points to extending HDC-based SQL operators to broader workloads.

Abstract

Hyperdimensional Computing (HDC) encodes information and data into high-dimensional distributed vectors that can be manipulated using simple bitwise operations and similarity searches, offering parallelism, low-precision hardware friendliness, and strong robustness to noise. These properties are a natural fit for SQL database workloads dominated by predicate evaluation and scans, which demand low energy and low latency over large fact tables. Notably, HDC's noise-tolerance maps well onto emerging ferroelectric NAND (FeNAND) memories, which provide ultra-high density and in-storage compute capability but suffer from elevated raw bit-error rates. In this work, we propose HDDB, a hardware-software co-design that combines HDC with FeNAND multi-level cells (MLC) to perform in-storage SQL predicate evaluation and analytics with massive parallelism and minimal data movement. Particularly, we introduce novel HDC encoding techniques for standard SQL data tables and formulate predicate-based filtering and aggregation as highly efficient HDC operations that can happen in-storage. By exploiting the intrinsic redundancy of HDC, HDDB maintains correct predicate and decode outcomes under substantial device noise (up to 10% randomly corrupted TLC cells) without explicit error-correction overheads. Experiments on TPC-DS fact tables show that HDDB achieves up to 80.6x lower latency and 12,636x lower energy consumption compared to conventional CPU/GPU SQL database engines, suggesting that HDDB provides a practical substrate for noise-robust, memory-centric database processing.

HDDB: Efficient In-Storage SQL Database Search Using Hyperdimensional Computing on Ferroelectric NAND Flash

TL;DR

HDDB addresses the energy- and bandwidth-intensive predicate evaluation in SQL analytics over large fact tables by co-designing Hyperdimensional Computing with FeNAND in-storage processing. It encodes SQL tables into hypervectors and implements in-storage predicate evaluation and decoding using HDC primitives, achieving noise robustness and high parallelism. The system introduces a columnar, in-storage mapping, a DBAM-based approximate matching accelerator, and specialized near-storage peripherals, delivering up to 80.6× latency reduction and 12,636× energy savings over CPU/GPU baselines while tolerating up to 10% random TLC cell corruption. This work demonstrates a practical memory-centric database processing substrate and points to extending HDC-based SQL operators to broader workloads.

Abstract

Hyperdimensional Computing (HDC) encodes information and data into high-dimensional distributed vectors that can be manipulated using simple bitwise operations and similarity searches, offering parallelism, low-precision hardware friendliness, and strong robustness to noise. These properties are a natural fit for SQL database workloads dominated by predicate evaluation and scans, which demand low energy and low latency over large fact tables. Notably, HDC's noise-tolerance maps well onto emerging ferroelectric NAND (FeNAND) memories, which provide ultra-high density and in-storage compute capability but suffer from elevated raw bit-error rates. In this work, we propose HDDB, a hardware-software co-design that combines HDC with FeNAND multi-level cells (MLC) to perform in-storage SQL predicate evaluation and analytics with massive parallelism and minimal data movement. Particularly, we introduce novel HDC encoding techniques for standard SQL data tables and formulate predicate-based filtering and aggregation as highly efficient HDC operations that can happen in-storage. By exploiting the intrinsic redundancy of HDC, HDDB maintains correct predicate and decode outcomes under substantial device noise (up to 10% randomly corrupted TLC cells) without explicit error-correction overheads. Experiments on TPC-DS fact tables show that HDDB achieves up to 80.6x lower latency and 12,636x lower energy consumption compared to conventional CPU/GPU SQL database engines, suggesting that HDDB provides a practical substrate for noise-robust, memory-centric database processing.

Paper Structure

This paper contains 17 sections, 3 equations, 8 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background and Related Works
  3. HDDB Algorithm Design
  4. Mapping SQL Table into Hypervectors
  5. Predicate Evaluation & Decode
  6. HDDB In-Storage Accelerator
  7. System Overview
  8. In-situ Approximate Matching and Peripherals Design
  9. HDDB In-Storage Mapping and Scheduling
  10. Evaluation
  11. HDDB Algorithm Evaluation
  12. HDDB In-Storage Accelerator Evaluation
  13. Experimental Setup
  14. Comparison with SOTA Database Engine
  15. Area and Power Analysis
  16. ...and 2 more sections

Figures (8)

  • Figure 1: End-to-end HDDB predicate evaluation pipeline and acceleration targets
  • Figure 2: Mapping (a) String data (b) Numeric data into HV
  • Figure 3: Overview of HDDB/ISP (a) Reconfigurable H-tree connected architecture (b) Predicate evaluation dataflow
  • Figure 4: FeNAND-based Dual Boundary Approximate Matching (DBAM) (a) 3D FeNAND array structure (b) 3D FeNAND planes (c) DBAM matched case (d) DBAM unmatched case
  • Figure 5: Design of (a) Heterogeneous integration (b) Encoded table near-storage processor (NSP) (c) LU dictionary NSP
  • ...and 3 more figures