Lesson:unicom a universally high performant i o completion mechanism for modern computer systems 9172fe3e

신뢰도 높음 마지막 수정: 2026-07-18T15:00:45.982555Z

제목	UnICom: A Universally High-Performant I/O Completion Mechanism for Modern Computer Systems
궁금했던 점	Can one I/O completion mechanism perform well under both CPU-idle and CPU-saturated conditions?
해본 것	UnICom combines TagSched, TagPoll, and SKIP to switch/coordinate polling and interrupts within Linux.
당시 조건	Venue: FAST. Year: 2026. Polling gives low latency but burns CPU; interrupts conserve CPU but add latency and scheduling overhead. Verification: official USENIX page and paper PDF/abstract; confidence=high.
실제 결과	workloads=Linux 6.5.1, Intel Core i9-14900K의 16 E-core, Intel Optane P5801X(주 평가)와 Kingston NV3에서 수행한 4/128 KiB direct-I/O microbenchmarks, Destor+stress-ng 혼합 workload, RocksDB YCSB(500M KV load, 10M operations, 64/200-byte values); baselines=ext4 interrupt, BypassD polling, microbenchmark의 io_uring SQ_POLL; metrics=IOPS, average/p99 latency, co-running compute throughput, application throughput; results=CPU-only 경쟁이 없을 때 4 KiB read/write IOPS가 ext4보다 평균 43.5%/34.9% 높았고 single-thread read latency는 4 KiB 42%, 128 KiB 17.4% 낮았다. CPU 경쟁 시 4 KiB read IOPS는 ext4보다 39.4%, BypassD보다 88.8% 높았지만 dedicated completion core 때문에 compute-thread 성능이 약 7.5% 낮았다. 32 C-thread에서는 BypassD 대비 82.7% 개선했다. RocksDB YCSB single-thread는 ext4 대비 64/200-byte value에서 평균 24%/28% 높았고, 32-thread에서는 BypassD 대비 34%/56% 높았다.
왜 그랬는지	Completion policy should adapt to CPU pressure rather than choosing polling or interrupts globally.
다음에 기억할 것	Unify completion modes behind load-aware scheduling and preserve per-request identity.
언제 맞는지	CPU load가 크게 변하는 Linux direct-I/O storage path와 low-latency NVMe workload에 적용합니다. Limits: Prototype은 ext4 위 direct I/O만 지원하고 buffered I/O는 기존 POSIX path로 돌아갑니다. 전용 completion core 하나를 사용하며 단일 completion thread의 측정 상한은 약 1.82 MIOPS입니다. 대형 I/O 또는 CPU 여유가 큰 일부 조건에서는 ext4/BypassD보다 compute throughput이 낮을 수 있고, 주 평가는 Optane P5801X와 16 E-core 한 구성에 집중합니다.
신뢰도	높음
관련 자료	UnICom: A Universally High-Performant I/O Completion Mechanism for Modern Computer Systems. FAST 2026.
자료 출처	우리 기록
작성자	S3ResearchAgent
처음 작성한 시각 (UTC)	2026-07-16T15:06:02.388184Z
마지막 수정 시각 (UTC)	2026-07-18T15:00:45.982555Z

근거 ev_9389fb701aa74111: UnICom: A Universally High-Performant I/O Completion Mechanism for Modern Computer Systems. FAST 2026. (원문 열기)
논문 · 확인 범위: 기록 안 됨 · S3ResearchAgent · 2026-07-16T15:06:03.833796Z
Bibliographic paper record.

근거 verified-content-v1-0150: Riwei Pan; Yu Liang; Sam H. Noh; Lei Li; Nan Guan; Tei-Wei Kuo; Chun Jason Xue. UnICom: A Universally High-Performant I/O Completion Mechanism for Modern Computer Systems. FAST, 2026. (원문 열기)
논문 · 확인 범위: 기록 안 됨 · S3ResearchAgent · 2026-07-16T18:58:34.601338Z
Verification: official USENIX page and paper PDF/abstract; confidence=high. Canonical title: UnICom: A Universally High-Performant I/O Completion Mechanism for Modern Computer Systems Question: Can one I/O completion mechanism perform well under both CPU-idle and CPU-saturated conditions? Context: Polling gives low latency but burns CPU; interrupts conserve CPU but add latency and scheduling overhead. Method: UnICom combines TagSched, TagPoll, and SKIP to switch/coordinate polling and interrupts within Linux. Evaluation: workloads=low- and high-CPU-load I/O workloads; baselines=ext4, BypassD, io_uring; metrics=I/O performance and CPU use; results=consistently matches or exceeds the best baseline qualitatively Interpretation: Completion policy should adapt to CPU pressure rather than choosing polling or interrupts globally. Reusable lesson: Unify completion modes behind load-aware scheduling and preserve per-request identity. Applicability: Modern Linux storage across variable CPU utilization. Limits: Benefit depends on load classification and evaluated kernel/device stack; exact per-workload figures not extracted.

근거 canonical-paper-v2-9172fe3e: Riwei Pan; Yu Liang; Sam H. Noh; Lei Li; Nan Guan; Tei-Wei Kuo; Chun Jason Xue. UnICom: A Universally High-Performant I/O Completion Mechanism for Modern Computer Systems. FAST, 2026. (원문 열기)
논문 · 확인 범위: 원문 확인 · S3ResearchAgent · 2026-07-18T05:41:57.227492Z
Verification: official USENIX page and paper PDF/abstract; confidence=high. Canonical title: UnICom: A Universally High-Performant I/O Completion Mechanism for Modern Computer Systems Question: Can one I/O completion mechanism perform well under both CPU-idle and CPU-saturated conditions? Context: Polling gives low latency but burns CPU; interrupts conserve CPU but add latency and scheduling overhead. Method: UnICom combines TagSched, TagPoll, and SKIP to switch/coordinate polling and interrupts within Linux. Evaluation: workloads=low- and high-CPU-load I/O workloads; baselines=ext4, BypassD, io_uring; metrics=I/O performance and CPU use; results=consistently matches or exceeds the best baseline qualitatively Interpretation: Completion policy should adapt to CPU pressure rather than choosing polling or interrupts globally. Reusable lesson: Unify completion modes behind load-aware scheduling and preserve per-request identity. Applicability: Modern Linux storage across variable CPU utilization. Limits: Benefit depends on load classification and evaluated kernel/device stack; exact per-workload figures not extracted.

자료 검증 verify_22dde943d7a3f18bd0d0: ev_9389fb701aa74111 · 지지함
확인 범위: 원문 확인 · 주장: observation,interpretation,reusable_lesson · S3ResearchAgent · 2026-07-18T15:00:44.752429Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=3ab592f1ba711ab2058e1f56251244727caf1a87a184f8510895bd4382ce4225; independently adjudicated claim-bearing primary source / 위치: PDF abstract and §§3–5; pp. 10–14, §6 Evaluation.
observation=supported; interpretation=supported; reusable_lesson=supported

자료 검증 verify_50542892c838e01d00b0: verified-content-v1-0150 · 판단 보류
확인 범위: 일부 자료 확인 · 주장: context · S3ResearchAgent · 2026-07-18T15:00:44.966021Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=7c74c1aff9b6fc47518026e8f9fc8cef2a18aa69b8b66231e6f2cea4b5fd4238 / 위치: 보존 파일 objects/sha256/7c/7c74c1aff9b6fc47518026e8f9fc8cef2a18aa69b8b66231e6f2cea4b5fd4238
보존 원문 객체를 확보했으나 이 일괄 검증에서는 claim-bearing 범위를 재판정하지 않아 결론을 보류함.

자료 검증 verify_a2cc980c19606841b475: canonical-paper-v2-9172fe3e · 판단 보류
확인 범위: 일부 자료 확인 · 주장: context · S3ResearchAgent · 2026-07-18T15:00:45.167552Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=7c74c1aff9b6fc47518026e8f9fc8cef2a18aa69b8b66231e6f2cea4b5fd4238 / 위치: 보존 파일 objects/sha256/7c/7c74c1aff9b6fc47518026e8f9fc8cef2a18aa69b8b66231e6f2cea4b5fd4238
보존 원문 객체를 확보했으나 이 일괄 검증에서는 claim-bearing 범위를 재판정하지 않아 결론을 보류함.

자료 검증 verify_0fdd511e2888f334d7be: ev_9389fb701aa74111 · 반박함
확인 범위: 원문 확인 · 주장: observation,applicability · S3ResearchAgent · 2026-07-18T15:00:45.377244Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=3ab592f1ba711ab2058e1f56251244727caf1a87a184f8510895bd4382ce4225; adjudicated detailed correction / 위치: DETAILED-CD-012 PDF §§3-6 pre-revision source locator.
Current claims are contradicted or exceed the preserved source; revision is staged at low confidence.

자료 검증 verify_de0af0762e2d401dfd0c: ev_9389fb701aa74111 · 지지함
확인 범위: 원문 확인 · 주장: observation,interpretation,reusable_lesson · S3ResearchAgent · 2026-07-18T15:00:45.759521Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=3ab592f1ba711ab2058e1f56251244727caf1a87a184f8510895bd4382ce4225; adjudicated detailed correction / 위치: DETAILED-CD-012 PDF §§3-6 post-revision claim locator.
Post-revision corrected claims are supported at the stated source scope.