ADDP: The Data Prefetching Protocol for Monitoring Capacity Misses
Open Access
- 11 April 2025
- journal article
- Published by Ediciones Universidad de Salamanca in ADCAIJ: Advances in Distributed Computing and Artificial Intelligence Journal
- Vol. 14, e31782
Abstract
Prefetching is essential to minimizing the number of misses in cache and improving processor performance. Many prefetchers have been proposed, including simple but highly effective stream-based prefetchers and prefetchers that predict complex access patterns based on structures such as history buffers and bit vectors. However, many cache misses still occur in many applications. After analyzing the various techniques in Instruction and Data Prefetcher, several key features were extracted which impact system performance. Data prefetching is an essential technique used in all commercial processors. Data prefetchers aim at hiding the long data access latency. In this paper, we present the design of an Adaptive Delta-based Data Prefetching (ADDP) that employs four different tables organized in a hierarchical manner to address the diversity of access patterns. Firstly, the Entry Table is queue, which tracks recent cache fill. Secondly, the Predict Table which has trigger (Program Counter) PCs as tags. Thirdly, the (Address Difference Table) ADT which has target PCs as tags. Lastly, the Prefetch Table is divided into two parts, i.e., Prefetch Filter and the actual Prefetch Table. The Prefetch Filter table filters unnecessary prefetch accesses and the Prefetch Table is used to track other additional information for each prefetch. The ADDP has been implemented in a multicache-level prefetching system under the 3rd Data Prefetching Championship (DPC-3) framework. ADDP is an effective solution for data-intensive applications since it shows notable gains in cache hit rates and latency reduction. The simulation results show that ADDP outperforms the top three data prefetchers MLOP, SPP and BINGO by 5.312 %, 13.213 % and 10.549 %, respectively.Keywords
This publication has 35 references indexed in Scilit:
- Efficient Flow Table Caching Architecture and Replacement Policy for SDN SwitchesJournal of Network and Systems Management, 2024
- Reinforcement Learning-Based Cache Replacement Policies for Multicore ProcessorsIEEE Access, 2024
- Deep learning based data prefetching in CPU-GPU unified virtual memoryJournal of Parallel and Distributed Computing, 2022
- Bouquet of Instruction Pointers: Instruction Pointer Classifier-based Spatial Hardware Prefetching2020
- A Survey of Computer Architecture Simulation Techniques and ToolsIEEE Access, 2019
- Analysis of Cyber Related Issues by Implementing Data Mining Algorithm2019
- Sandbox Prefetching: Safe run-time evaluation of aggressive prefetchers2014
- Characterization and dynamic mitigation of intra-application cache interference2011
- A Taxonomy of Data Prefetching Mechanisms2008
- Feedback Directed Prefetching: Improving the Performance and Bandwidth-Efficiency of Hardware Prefetchers2007