{"id":842,"date":"2021-06-21T18:00:36","date_gmt":"2021-06-21T09:00:36","guid":{"rendered":"https:\/\/wise.ajou.ac.kr:9605\/?page_id=842"},"modified":"2022-03-18T13:31:46","modified_gmt":"2022-03-18T04:31:46","slug":"%ec%97%ac%ec%83%81%ed%98%b8","status":"publish","type":"page","link":"https:\/\/wise.ajou.ac.kr\/?page_id=842","title":{"rendered":"\uc5ec\uc0c1\ud638"},"content":{"rendered":"\n
\"\"<\/figure><\/div>\n\n\n\n

\uc5ec\uc0c1\ud638
Sangho Yeo<\/sup><\/h1>\n\n\n\n

Email<\/h2>\n\n\n\n

soboru963 at ajou.ac.kr<\/p>\n\n\n\n

Research interests<\/h2>\n\n\n\n

GPGPU, Cloud<\/p>\n\n\n\n

Introduction<\/h2>\n\n\n\n

\uc548\ub155\ud558\uc138\uc694 \uc800\ub294 \uc544\uc8fc\ub300\ud559\uad50\ub97c \uc878\uc5c5\ud558\uace0 \uc544\uc8fc\ub300\ud559\uc6d0\uc5d0 \uc654\uc2b5\ub2c8\ub2e4. \ubd84\uc0b0\ud504\ub85c\uadf8\ub798\ubc0d, \ubcd1\ub82c\ud504\ub85c\uadf8\ub798\ubc0d\uc5d0 \ub300\ud574\uc11c \ucc98\uc74c \uc811\ud558\uac8c\ub41c \uacc4\uae30\ub294 \ub300\ud559\uad50\uc5d0\uc11c \ubc30\uc6b4 \ud558\ub461\uc785\ub2c8\ub2e4. \uadf8 \uc774\ud6c4\ub85c \ud074\ub77c\uc6b0\ub4dc \uc11c\ube44\uc2a4\uc778 AWS\ub97c \ud65c\uc6a9\ud558\uc5ec \ud559\uad50\uc5d0\uc11c \uac8c\uc784\ud50c\ub7ab\ud3fc \ud504\ub85c\uc81d\ud2b8\ub97c \ud574\ubcf4\uace0 \uac19\uc774 \ud504\ub85c\uc81d\ud2b8\ub97c \ud55c \ud615\uc758 \uc870\uc5b8\uc73c\ub85c \uc774 \uc5f0\uad6c\uc2e4\uc5d0 \uc624\uac8c\ub418\uc5c8\uc2b5\ub2c8\ub2e4. \uc774 \uc804\uacf5\uc774 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \ud65c\uc6a9\ub420 \uc218 \uc788\ub294 \uac83\ucc98\ub7fc \uc800\ub3c4 \ub300\ud559\uc6d0 \uc0dd\ud65c\uc744 \ud558\uba74\uc11c \ub2e4\uc591\ud55c \uacbd\ud5d8\uc744 \ud558\uba74 \uc88b\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n\n\n\n

Publications<\/h2>\n\n\n
<\/a><\/form>

2026<\/h3>
11.<\/div>

Yeo, Sangho; Kim, Jiheon; Oh, Sangyoon<\/p>

SpaceFlow: Runtime Enforcement of Cloud-Native GPU Spatial Partitioning<\/a> Journal Article<\/span> <\/p>

In: <\/span>Journal of Parallel and Distributed Computing, <\/span>2026<\/span>.<\/p>

Abstract<\/a><\/span> | Links<\/a><\/span> | BibTeX<\/a><\/span><\/p>

@article{sanghoyeojpdc202607,
\r\ntitle = {SpaceFlow: Runtime Enforcement of Cloud-Native GPU Spatial Partitioning},
\r\nauthor = {Sangho Yeo and Jiheon Kim and Sangyoon Oh },
\r\nurl = {https:\/\/www.sciencedirect.com\/science\/article\/pii\/S0743731526000894?__cf_chl_f_tk=9290J7qEI.ciFqQ7qLcwHCGVUby21dj2aNeWnhHOCUg-1783310840-1.0.1.1-0gj8vTrE9ZmcMFuOTF8xbgHcjhwMUo2Zt7mEhGVfTmA},
\r\ndoi = {https:\/\/doi.org\/10.1016\/j.jpdc.2026.105311},
\r\nyear  = {2026},
\r\ndate = {2026-07-02},
\r\nurldate = {2026-07-02},
\r\njournal = {Journal of Parallel and Distributed Computing},
\r\nabstract = {As GPU capacity scales, a growing fraction of computing resources remains idle under single-tenant allocations. Spatial partitioning addresses this need by offering superior throughput. However, it lacks container-level enforcement, causing resource over-consumption in multi-tenant environments. We present SpaceFlow, a runtime enforcement mechanism that achieves container-level GPU spatial partitioning by transparently intercepting GPU runtime interface and unifying its contexts. To ensure secure deployment, SpaceFlow integrates with a cloud-native ecosystem and eliminates resource oversubscription by immutable resource configuration. For broad compatibility, SpaceFlow includes adaptive fallback mechanisms that maintain container partitioning on GPUs with the latest architecture features. We validate SpaceFlow across various GPU types in a cloud-native cluster, demonstrating significant reduction in cross-container interference compared with conventional process-level spatial partitioning. SpaceFlow achieves up to 11.34\u202f\u00d7\u202f interference reduction in evaluated scenarios while maintaining \u202f\u2264\u202f3.0% overhead.},
\r\nkeywords = {},
\r\npubstate = {published},
\r\ntppubtype = {article}
\r\n}
\r\n<\/pre><\/div>
Close<\/a><\/p><\/div>
As GPU capacity scales, a growing fraction of computing resources remains idle under single-tenant allocations. Spatial partitioning addresses this need by offering superior throughput. However, it lacks container-level enforcement, causing resource over-consumption in multi-tenant environments. We present SpaceFlow, a runtime enforcement mechanism that achieves container-level GPU spatial partitioning by transparently intercepting GPU runtime interface and unifying its contexts. To ensure secure deployment, SpaceFlow integrates with a cloud-native ecosystem and eliminates resource oversubscription by immutable resource configuration. For broad compatibility, SpaceFlow includes adaptive fallback mechanisms that maintain container partitioning on GPUs with the latest architecture features. We validate SpaceFlow across various GPU types in a cloud-native cluster, demonstrating significant reduction in cross-container interference compared with conventional process-level spatial partitioning. SpaceFlow achieves up to 11.34\u202f\u00d7\u202f interference reduction in evaluated scenarios while maintaining \u202f\u2264\u202f3.0% overhead.<\/div>
Close<\/a><\/p><\/div>