{"id":1715,"date":"2022-03-30T15:53:22","date_gmt":"2022-03-30T06:53:22","guid":{"rendered":"https:\/\/wise.ajou.ac.kr\/?page_id=1715"},"modified":"2025-10-17T13:36:33","modified_gmt":"2025-10-17T04:36:33","slug":"%ec%b5%9c%ec%a7%80%ed%97%8c","status":"publish","type":"page","link":"https:\/\/wise.ajou.ac.kr\/?page_id=1715","title":{"rendered":"\ucd5c\uc9c0\ud5cc"},"content":{"rendered":"\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t
\"\uc775\uba85<\/figure>\n

<\/p>\n

<\/p>\n

\ucd5c\uc9c0\ud5cc
Jiheon Choi<\/sup><\/h1>\n

<\/p>\n

<\/p>\n

\n
\n

<\/p>\n

\n
\u00a0<\/div>\n<\/div>\n<\/div>\n<\/div>\n

<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t
\n\t\t\t\t\t\n\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\tHomepage<\/span>\n\t\t\t\t\t<\/span>\n\t\t\t\t\t<\/a>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t
\n
\n
\n
\n
\u00a0<\/div>\n<\/div>\n<\/div>\n

<\/p>\n<\/div>\n<\/div>\n

<\/p>\n

<\/p>\n

Email<\/h2>\n

<\/p>\n

<\/p>\n

unidev at\u00a0ajou.ac.kr<\/p>\n

<\/p>\n

<\/p>\n

Research interests<\/h2>\n

<\/p>\n

<\/p>\n

Distributed Systems, Secure Infrastructure for AI, Cloud Computing Security<\/p>\n

<\/p>\n

<\/p>\n

Introduction<\/h2>\n

<\/p>\n

<\/p>\n

Jiheon Choi is a Ph.D. student at Ajou University. he is interested in HPC, Distributed Systems, and MLOps.\u00a0<\/p>\n

<\/p>\n

<\/p>\n

Publications<\/h2>\n

<\/p>\n

<\/p>\n

<\/a><\/form>

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

Choi, Jiheon; Oh, Sangyoon<\/p>

S-CQR: Stratified Calibration for Runtime Prediction in HPC Backfill Scheduling Conference<\/span> Forthcoming<\/span><\/p>

The 32nd International European Conference on Parallel and Distributed Computing (Euro-Par 2026), <\/span>Forthcoming.<\/p>

BibTeX<\/a><\/span><\/p>

@conference{jiheon2026scqr,
\r\ntitle = {S-CQR: Stratified Calibration for Runtime Prediction in HPC Backfill Scheduling},
\r\nauthor = {Jiheon Choi and Sangyoon Oh},
\r\nyear  = {2026},
\r\ndate = {2026-08-24},
\r\nbooktitle = {The 32nd International European Conference on Parallel and Distributed Computing (Euro-Par 2026)},
\r\nkeywords = {},
\r\npubstate = {forthcoming},
\r\ntppubtype = {conference}
\r\n}
\r\n<\/pre><\/div>
Close<\/a><\/p><\/div><\/div><\/div>
16.<\/div>
 \uc815\ud601\uc900,;  \ucd94\ubbfc\uc194,;  \ucd5c\uc9c0\ud5cc,;  \uc624\uc0c1\uc724,<\/p>
LLM \uc11c\ube59 \uc2dc\uc2a4\ud15c\uc758 \uc694\uccad \uc720\ud615 \uac04 \uc790\uc6d0 \uacbd\ud569 \ubd84\uc11d\uacfc \uc720\ud615 \uae30\ubc18 \uc2a4\ucf00\uc904\ub9c1 Conference<\/span> Forthcoming<\/span><\/p>
2026\ub144\ub3c4 \ud55c\uad6d\uc815\ubcf4\uacfc\ud559\ud68c \ud55c\uad6d\ucef4\ud4e8\ud130\uc885\ud569\ud559\uc220\ub300\ud68c (KCC 2026), <\/span>Forthcoming.<\/p>
BibTeX<\/a><\/span><\/p>
@conference{hyeokjunkcc2026,
\r\ntitle = {LLM \uc11c\ube59 \uc2dc\uc2a4\ud15c\uc758 \uc694\uccad \uc720\ud615 \uac04 \uc790\uc6d0 \uacbd\ud569 \ubd84\uc11d\uacfc \uc720\ud615 \uae30\ubc18 \uc2a4\ucf00\uc904\ub9c1},
\r\nauthor = {\uc815\ud601\uc900 and \ucd94\ubbfc\uc194 and \ucd5c\uc9c0\ud5cc and \uc624\uc0c1\uc724 },
\r\nyear  = {2026},
\r\ndate = {2026-06-24},
\r\nurldate = {2026-06-24},
\r\nbooktitle = {2026\ub144\ub3c4 \ud55c\uad6d\uc815\ubcf4\uacfc\ud559\ud68c \ud55c\uad6d\ucef4\ud4e8\ud130\uc885\ud569\ud559\uc220\ub300\ud68c (KCC 2026)},
\r\nkeywords = {},
\r\npubstate = {forthcoming},
\r\ntppubtype = {conference}
\r\n}
\r\n<\/pre><\/div>
Close<\/a><\/p><\/div><\/div><\/div>
15.<\/div>
 Choi, Jiheon;  Oh, Sangyoon<\/p>
Reducing Backfill Failures with Lightweight Uncertainty Buffers under Workload Drift in HPC Job Scheduling Conference<\/span> <\/p>
The 26th IEEE\/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGrid 2026), <\/span>2026<\/span>.<\/p>
BibTeX<\/a><\/span><\/p>
@conference{CCGrid-2026,
\r\ntitle = {Reducing Backfill Failures with Lightweight Uncertainty Buffers under Workload Drift in HPC Job Scheduling},
\r\nauthor = {Jiheon Choi and Sangyoon Oh},
\r\nyear  = {2026},
\r\ndate = {2026-05-21},
\r\nurldate = {2026-05-21},
\r\nbooktitle = {The 26th IEEE\/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGrid 2026)},
\r\nkeywords = {},
\r\npubstate = {published},
\r\ntppubtype = {conference}
\r\n}
\r\n<\/pre><\/div>
Close<\/a><\/p><\/div><\/div><\/div>
14.<\/div>
 Choi, Jiheon;  Oh, Sangyoon<\/p>
UARP: Uncertainty-aware runtime prediction for preventing scheduler termination under Wallclock constraints in HPC<\/a> Journal Article<\/span> <\/p>
In: <\/span>The Journal of Supercomputing, <\/span>vol. 82, <\/span>2026<\/span>.<\/p>
Abstract<\/a><\/span> | Links<\/a><\/span> | BibTeX<\/a><\/span><\/p>
@article{jiheon2026uarp,
\r\ntitle = {UARP: Uncertainty-aware runtime prediction for preventing scheduler termination under Wallclock constraints in HPC},
\r\nauthor = {Jiheon Choi and Sangyoon Oh},
\r\nurl = {https:\/\/link.springer.com\/article\/10.1007\/s11227-026-08422-8},
\r\nyear  = {2026},
\r\ndate = {2026-03-25},
\r\nurldate = {2026-03-25},
\r\njournal = {The Journal of Supercomputing},
\r\nvolume = {82},
\r\nabstract = {Effective resource allocation has become a critical issue in high-performance computing (HPC) systems. To effectively allocate resources (e.g., CPU\/GPU cores), recent studies focus on predicting each workload\u2019s runtime using machine learning and deep learning models. These methods in HPC often suffer from underestimation, as 33\u201364% of jobs terminate due to wallclock time limits, whereas user-provided estimates achieve 78\u201399% success. This failure stems from minimizing mean squared error, which biases predictions toward average-case performance and underestimates jobs in high-skewed (i.e., long-tail) runtime distributions. Specifically, HPC workloads exhibit long-tail runtime distributions, with most jobs completing quickly while a small fraction runs for extremely long durations. To overcome these challenges, we introduce an uncertainty-aware runtime prediction (UARP) method based on multi-quantile regression. Our method directly addresses the underestimation problem by quantifying uncertainty by modeling the conditional distribution without distributional assumptions. Our approach uses the highest-quantile (99th) model and the residual model from the median quantile. The 99th model primarily provides conservative bounds and protection against job underestimates, while the residual uncertainty model protects against unpredictable workloads by estimating prediction variance. In particular, the expected predicted error (i.e., uncertainty) from the residual model plays a critical role in our adaptive safety margin calculation. UARP adds a conservative prediction (99th quantile) and an additional safety margin from our formula, enabling our adaptive margin approach specifically tailored to each job\u2019s characteristics. Evaluation on four production HPC systems (SDSC DataStar, KIT FH2, ANL Interpid, KISTI NURION), UARP achieves 92\u201399% job success rates while maintaining resource utilization within 1\u20132% of EASY backfilling. Our method deploys with identical parameters across all systems. This parameter-free deployment eliminates the per-system tuning that fixed-margin approaches require. In addition, our approach integrates with existing schedulers through minimal modification, utilizing uncertainty-aware predictions to prevent timeout-based job termination and preserve system efficiency.},
\r\nkeywords = {},
\r\npubstate = {published},
\r\ntppubtype = {article}
\r\n}
\r\n<\/pre><\/div>
Close<\/a><\/p><\/div>
Effective resource allocation has become a critical issue in high-performance computing (HPC) systems. To effectively allocate resources (e.g., CPU\/GPU cores), recent studies focus on predicting each workload\u2019s runtime using machine learning and deep learning models. These methods in HPC often suffer from underestimation, as 33\u201364% of jobs terminate due to wallclock time limits, whereas user-provided estimates achieve 78\u201399% success. This failure stems from minimizing mean squared error, which biases predictions toward average-case performance and underestimates jobs in high-skewed (i.e., long-tail) runtime distributions. Specifically, HPC workloads exhibit long-tail runtime distributions, with most jobs completing quickly while a small fraction runs for extremely long durations. To overcome these challenges, we introduce an uncertainty-aware runtime prediction (UARP) method based on multi-quantile regression. Our method directly addresses the underestimation problem by quantifying uncertainty by modeling the conditional distribution without distributional assumptions. Our approach uses the highest-quantile (99th) model and the residual model from the median quantile. The 99th model primarily provides conservative bounds and protection against job underestimates, while the residual uncertainty model protects against unpredictable workloads by estimating prediction variance. In particular, the expected predicted error (i.e., uncertainty) from the residual model plays a critical role in our adaptive safety margin calculation. UARP adds a conservative prediction (99th quantile) and an additional safety margin from our formula, enabling our adaptive margin approach specifically tailored to each job\u2019s characteristics. Evaluation on four production HPC systems (SDSC DataStar, KIT FH2, ANL Interpid, KISTI NURION), UARP achieves 92\u201399% job success rates while maintaining resource utilization within 1\u20132% of EASY backfilling. Our method deploys with identical parameters across all systems. This parameter-free deployment eliminates the per-system tuning that fixed-margin approaches require. In addition, our approach integrates with existing schedulers through minimal modification, utilizing uncertainty-aware predictions to prevent timeout-based job termination and preserve system efficiency.<\/div>
Close<\/a><\/p><\/div>