]> 2026-09-08T16:38:33+09:00 Lesson:technical review exploiting gpus in virtual machine for biocloud 9e8875c0 3000 ko 장시간 GPU batch/HPC job을 호스팅하는 VM 환경에 해당하며, fine-grained multi-tenant inference에는 그대로 일반화할 수 없다. Xen에서 PCIe direct pass-through로 GPU를 VM에 직접 연결하고, PCIe hot plug-in/out과 GPU-Admin/GPU-Manager로 workload 단위 coarse-grained sharing을 수행한다. S3ResearchAgent high Verification level: full_text. Sources checked: https://doi.org/10.1155/2013/939460 https://pmc.ncbi.nlm.nih.gov/articles/PMC3654629/ https://pubmed.ncbi.nlm.nih.gov/23710465/ 2026-07-16T18:51:17Z 2461238.2856134 Verification source: https://doi.org/10.1155/2013/939460. Claims are limited to the stated access level; unverified details are explicitly marked. sharing 시나리오는 synthetic하고 효과는 workload 실행시간·idle time·GPU 수에 따라 달라진다. coarse-grained exclusivity라서 짧고 동시성이 높은 workload보다 장시간 bioinformatics workload에 맞는다. Technical interpretation: device passthrough의 near-native datapath와 hot-plug 기반 coarse-grained multiplexing을 분리하면 장시간 accelerator job에서 isolation·throughput·sharing을 함께 얻을 수 있다. Technical Review: Exploiting GPUs in Virtual Machine for BioCloud BlackScholes의 virtualization overhead는 0.5% 미만(비교 대상은 25–73%), 네 bioinformatics application의 전체 실행 overhead는 평균 3%, 최대 10%였다. hot plug-in/out은 각각 1.3±0.1초였다. 네 VM이 한 GPU를 공유할 때 idle/sleep 구간에 따라 총 실행시간이 20–53% 줄었다. 한 GPU일 때 평균 대기 25.05초, 두 GPU일 때 4.15초였다. bioinformatics처럼 긴 GPU workload에 대해 VM이 GPU를 near-native 성능으로 쓰면서 여러 VM이 GPU를 time-share할 수 있는가? lab device passthrough의 near-native datapath와 hot-plug 기반 coarse-grained multiplexing을 분리하면 장시간 accelerator job에서 isolation·throughput·sharing을 함께 얻을 수 있다. Draft 2026-07-18T15:00:36Z 2461240.1254167 2026-07-18T15:00:36Z 2461240.1254167 technical review exploiting gpus in virtual machine for biocloud 9e8875c0 3000 ev_593536927d28441c 2026-07-16T18:51:21Z 2461238.2856597 S3ResearchAgent Primary or official publication source used for the technical review. paper Verification level: full_text. Technical claims in this lesson are restricted to content exposed by this source. not_recorded technical review exploiting gpus in virtual machine for biocloud 9e8875c0# 5882ee1c2d862e4da4af8c57cfc99c21 3000 2026-07-18T15:00:36Z 2461240.1254167 S3ResearchAgent verify_992a48516b2669eb559f full_text observation,interpretation,reusable_lesson observation=supported; interpretation=supported; reusable_lesson=supported supports 9660f6d73e1256bbb59b71743e5b969dbeb1779882375723ea0afc47626ae05b R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=9660f6d73e1256bbb59b71743e5b969dbeb1779882375723ea0afc47626ae05b; independently adjudicated claim-bearing primary source Sections 3.1-3.3 and 4.1-4.3; Figures 5-8; Tables 2-3. ev_593536927d28441c e3e1dc549b2e615afe5cd2bfb91868a243cada1fab535e649501d0edb12442d4 technical review exploiting gpus in virtual machine for biocloud 9e8875c0# 4d44964bd22da2a65f2bca3586ab15f7