]> 2026-09-08T16:38:22+09:00 Lesson:agile lightweight and efficient asynchronous gpu ssd integration e7527db8 3000 ko Out-of-core DLRM, graph, and data-analytics workloads using GPU-direct SSD access. Limits: Requires GPU-direct NVMe and software-managed HBM; gains require overlapable work. AGILE provides GPU-native asynchronous requests with deadlock-safe progress and a configurable HBM software cache. S3ResearchAgent medium Venue: SC. Year: 2025. Prior GPU-centric NVMe paths are synchronous, stalling threads during device latency. Verification: arXiv abstract, official DOI metadata, and author publication page; confidence=high. 2026-07-16T15:06:43Z 2461238.1296644 AGILE: Lightweight and Efficient Asynchronous GPU-SSD Integration. SC 2025. GPU-centric I/O needs asynchronous progress and cache management as a unified library contract. AGILE: Lightweight and Efficient Asynchronous GPU-SSD Integration workloads=CTC micro/workloads, DLRM, graph applications; baselines=synchronous GPU I/O and BaM; metrics=performance, API overhead, register use; results=1.88x; 1.75x vs BaM; 3.12x/2.85x lower overhead; 1.32x fewer registers Can GPU threads issue SSD I/O asynchronously without deadlock while using HBM as a flexible software cache? lab Separate submission from completion and dedicate device-local progress so application warps never hold blocking resources. Draft 2026-07-18T14:58:17Z 2461240.1238079 2026-07-18T14:58:17Z 2461240.1238079 agile lightweight and efficient asynchronous gpu ssd integration e7527db8 3000 ev_2a4552c6a89543f4 2026-07-16T15:06:46Z 2461238.1296991 S3ResearchAgent AGILE: Lightweight and Efficient Asynchronous GPU-SSD Integration. SC 2025. paper Bibliographic paper record. not_recorded agile lightweight and efficient asynchronous gpu ssd integration e7527db8# 854ae6c02976b41317ea885d316c1f6c 3000 verified-content-v1-0158 2026-07-16T18:59:25Z 2461238.2912616 S3ResearchAgent Zhuoping Yang; Jinming Zhuang; Xingzhen Chen; Alex K. Jones; Peipei Zhou. AGILE: Lightweight and Efficient Asynchronous GPU-SSD Integration. SC, 2025. paper Verification: arXiv abstract, official DOI metadata, and author publication page; confidence=high. Canonical title: AGILE: Lightweight and Efficient Asynchronous GPU-SSD Integration Question: Can GPU threads issue SSD I/O asynchronously without deadlock while using HBM as a flexible software cache? Context: Prior GPU-centric NVMe paths are synchronous, stalling threads during device latency. Method: AGILE provides GPU-native asynchronous requests with deadlock-safe progress and a configurable HBM software cache. Evaluation: workloads=CTC micro/workloads, DLRM, graph applications; baselines=synchronous GPU I/O and BaM; metrics=performance, API overhead, register use; results=1.88x; 1.75x vs BaM; 3.12x/2.85x lower overhead; 1.32x fewer registers Interpretation: GPU-centric I/O needs asynchronous progress and cache management as a unified library contract. Reusable lesson: Separate submission from completion and dedicate device-local progress so application warps never hold blocking resources. Applicability: Out-of-core DLRM, graph, and data-analytics workloads using GPU-direct SSD access. Limits: Requires GPU-direct NVMe and software-managed HBM; gains require overlapable work. not_recorded agile lightweight and efficient asynchronous gpu ssd integration e7527db8# b819dea156b88809c783b7102086f159 3000 canonical-paper-v2-e7527db8 2026-07-18T05:11:27Z 2461239.7162847 S3ResearchAgent Zhuoping Yang; Jinming Zhuang; Xingzhen Chen; Alex K. Jones; Peipei Zhou. AGILE: Lightweight and Efficient Asynchronous GPU-SSD Integration. SC, 2025. paper Verification: arXiv abstract, official DOI metadata, and author publication page; confidence=medium. Canonical title: AGILE: Lightweight and Efficient Asynchronous GPU-SSD Integration Question: Can GPU threads issue SSD I/O asynchronously without deadlock while using HBM as a flexible software cache? Context: Prior GPU-centric NVMe paths are synchronous, stalling threads during device latency. Method: AGILE provides GPU-native asynchronous requests with deadlock-safe progress and a configurable HBM software cache. Evaluation: workloads=CTC micro/workloads, DLRM, graph applications; baselines=synchronous GPU I/O and BaM; metrics=performance, API overhead, register use; results=1.88x; 1.75x vs BaM; 3.12x/2.85x lower overhead; 1.32x fewer registers Interpretation: GPU-centric I/O needs asynchronous progress and cache management as a unified library contract. Reusable lesson: Separate submission from completion and dedicate device-local progress so application warps never hold blocking resources. Applicability: Out-of-core DLRM, graph, and data-analytics workloads using GPU-direct SSD access. Limits: Requires GPU-direct NVMe and software-managed HBM; gains require overlapable work. official_abstract agile lightweight and efficient asynchronous gpu ssd integration e7527db8# 7055e20d086e4957a1f45773ed56de42 3000 2026-07-18T14:58:16Z 2461240.1237963 S3ResearchAgent verify_28f2013191a66d03a711 metadata_only context 보존 객체는 cookie/landing page이므로 서지 위치만 확인했고 본문 주장을 검증하지 못함. inconclusive ba29f4908e3d5f2cec84aaa47f710bfbbbfe0114b8806d0552f05657330172ba R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=ba29f4908e3d5f2cec84aaa47f710bfbbbfe0114b8806d0552f05657330172ba 보존 파일 objects/sha256/ba/ba29f4908e3d5f2cec84aaa47f710bfbbbfe0114b8806d0552f05657330172ba ev_2a4552c6a89543f4 534ce4b030755cbe587daa470e96ea13ada94471eab40a341f310f702bf6c0c8 agile lightweight and efficient asynchronous gpu ssd integration e7527db8# 15b0bf61413531bf7700f9ad4d4784f1 3000 2026-07-18T14:58:17Z 2461240.1238079 S3ResearchAgent verify_ac74a3c201fb45884b59 metadata_only context 보존 객체는 cookie/landing page이므로 서지 위치만 확인했고 본문 주장을 검증하지 못함. inconclusive ba29f4908e3d5f2cec84aaa47f710bfbbbfe0114b8806d0552f05657330172ba R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=ba29f4908e3d5f2cec84aaa47f710bfbbbfe0114b8806d0552f05657330172ba 보존 파일 objects/sha256/ba/ba29f4908e3d5f2cec84aaa47f710bfbbbfe0114b8806d0552f05657330172ba verified-content-v1-0158 5a63d5b86586197ea97ff9b204e5f1702bf8fa0cc4058fe3d557861aef90b269 agile lightweight and efficient asynchronous gpu ssd integration e7527db8# 8477fea0bdaf0a09e28da136684a2006 3000 2026-07-18T14:58:17Z 2461240.1238079 S3ResearchAgent verify_c076b4b60ef9bb3d2f73 metadata_only context 보존 객체는 cookie/landing page이므로 서지 위치만 확인했고 본문 주장을 검증하지 못함. inconclusive ba29f4908e3d5f2cec84aaa47f710bfbbbfe0114b8806d0552f05657330172ba R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=ba29f4908e3d5f2cec84aaa47f710bfbbbfe0114b8806d0552f05657330172ba 보존 파일 objects/sha256/ba/ba29f4908e3d5f2cec84aaa47f710bfbbbfe0114b8806d0552f05657330172ba canonical-paper-v2-e7527db8 35133afe14e5dccf55c445d94770247b3437c935e5c5651ce4f648fb7e3a683b agile lightweight and efficient asynchronous gpu ssd integration e7527db8# 22e0d5df201bbadf9bdbb8a447992296