Researchers from the Division of Vitality’s Oak Ridge Nationwide Laboratory have developed a brand new utility to extend effectivity in reminiscence methods for high-performance computing.
Relatively than permit knowledge to bathroom down conventional reminiscence methods in supercomputers and influence efficiency, the workforce from ORNL, together with researchers from the College of Tennessee, Knoxville, created a framework to handle knowledge extra effectively with reminiscence methods that make use of extra advanced buildings. Analysis papers detailing their work have been just lately accepted in ACM Transactions on Structure and Code Optimization and the Worldwide Journal of Excessive-Efficiency Computing Functions.
Working underneath the Exascale Computing Challenge, or ECP, a multi-year software program analysis, growth and deployment venture managed by DOE, ORNL senior laptop science researcher Terry Jones and his workforce titled their work the “ECP Simplified Interface to Complicated Reminiscences,” or SICM, Challenge.
“Our work is to robotically put the continuously used objects into the proper location within the quicker tier of reminiscence and put the less-used objects, the issues that are not accessed as typically, into the slower reminiscence,” mentioned Jones. “Our work exhibits it performs higher than earlier methods.”
To optimize the huge quantities of knowledge saved on high-performance computer systems like ORNL’s Frontier, the world’s first exascale supercomputer, scientists want methods to construction reminiscence methods primarily based on the necessity for the saved data. Reminiscence methods that function quicker for data retrieval might be computationally costly in that they require extra computing energy to finish, whereas methods that include extra knowledge typically function at slower speeds.
Jones additionally mentioned that, conventionally, reminiscence methods have operated on a “first contact” precept, the place knowledge is saved within the quickest reminiscence storage till it reaches capability. Nevertheless, in lots of instances, the preliminary levels of a program embody components that can solely be used when this system is initiated, filling the quickest reminiscence areas with components that can not be wanted.
“First contact is a much less excellent strategy for these sorts of purposes,” Jones mentioned. “Our strategy makes use of extra refined strategies to find out if some knowledge wants quicker reminiscence or not and can provide you significantly better efficiency than first contact.”
Utilizing the SICM system, data is robotically sorted and saved primarily based on want, making retrieval considerably extra environment friendly and permitting builders to write down applications that higher use the total capability of supercomputing methods. Transferring ahead, this software will allow a number of applications with completely different storage must operate inside a single supercomputing rack by way of a brand new expertise referred to as CXL.
“Think about that inside a rack of a supercomputer there’s numerous reminiscence, and all of the nodes inside that very same rack may get no matter they want from that reminiscence,” Jones mentioned. “So, if inside a rack there are a number of applications, similar to an AI utility and a posh calculation on a small dataset, the AI program will want plenty of reminiscence, however the advanced calculation is not going to want as a lot reminiscence. Dynamically inside that rack, we may have this reminiscence transfer round whereas these two codes are operating.”
Extra data:
Brandon Kammerdiener et al, Versatile and Efficient Object Tiering for Heterogeneous Reminiscence Programs, ACM Transactions on Structure and Code Optimization (2024). DOI: 10.1145/3708540
Roxana Bujack et al, The ECP SICM venture: Managing advanced reminiscence hierarchies for exascale purposes, The Worldwide Journal of Excessive Efficiency Computing Functions (2024). DOI: 10.1177/10943420241288243
Quotation:
Supercomputing reminiscence administration software makes knowledge storage extra environment friendly (2025, March 26)
retrieved 26 March 2025
from https://techxplore.com/information/2025-03-supercomputing-memory-tool-storage-efficient.html
This doc is topic to copyright. Aside from any honest dealing for the aim of personal examine or analysis, no
half could also be reproduced with out the written permission. The content material is offered for data functions solely.
