Value in OptANNe: Accelerating Graph-based ANN with Optane Memory
Discover how OptANNe leverages Optane Persistent Memory to enhance Approximate Nearest Neighbors search for graph-based applications. It offers exceptional value across multiple axes, including accuracy, scaling, performance, and cost-efficiency. The future roadmap involves advancements in algorithms, hardware, and software to further optimize graph navigability, reduce DRAM usage, and scale solutions for larger datasets.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
OPtANNe: Optane Persistent Memory-based Approximate Nearest Neighbors Sourabh Dongaonkar1, Mark Hildebrand1,3, Mariano Tepper2, Cecilia Aguerrebere2, Ted Willke2, Jawad Khan1 1Intel Corporation, 2Intel Labs, 3UC Davis 1 Department or Event Name Intel Confidential
Optane Persistent Memory Accessed as memory on the DDR bus DIMM form factor High capacity (up to 512GB per DIMM) Nonvolatile memory based on 3DXP Optane occupies a previously empty position in the memory/storage hierarchy 2 Department or Event Name Intel Confidential
Why Optane for Graph-based ANN? Greedy graph search: query entry point 1. Retrieve list of neighbors of the current node. 2. Find the closest neighbor p to the query. nearest neighbor 3. Set current node = p Graph index Less frequent random accesses + large graph for high recall Data vectors Frequent random accesses of small chunks of data Graph HD vectors Speed Capacity Cost Place in Optane PMem Place in DRAM DRAM DRAM Optane DRAM Two main data structures, with different memory access patterns SSD DRAM Icons: "https://www.freepik.com" 3 Department or Event Name Intel Confidential
OptANNe offers exceptional value across all 5 key axes Accuracy A multifaceted problem needs multifaceted measures Scaling Performance ANN Value stemming from: Solid SW engineering Optane offers GB/$ to store large graphs Optane offers fast random load/store access Power Cost (CAPEX+OPEX) Improvement over next-best solution (CAPEX+OPEX) Deep1B BigANN MSTuring MSSpace Text2Image ~20x ~20x ~10x ~10x ~4x to ~9x 4 Department or Event Name Intel Confidential
Future work and call to action Next-gen algorithms Improve graph navigability Fewer hops for faster search Sparser graphs for BIGGER-ANN Reduce DRAM usage with better compact representations (PQ, ScaNN & new methods) Next-gen hardware More parallelism with more CPUs PMem pooling with CXL Next-gen software Even the algorithmic playing field with new SOTA SW Next-gen scaling Highly proficient solutions for 1B Extend the scope of future challenges to 10B, 100B, 1T 5 Department or Event Name Intel Confidential
