InternationalUnder Review2026
HyZNS: A Flexible Hybrid of Zoned and Conventional Namespaces in a ZNS SSD
Doeun Kim, Junmo Seong, Jinyoung Kim, Junill Jang, Hyuck Han, Sooyong Kang
Journal of Systems Architecture
SSD Firmware · Storage Systems
Junmo Seong
I work on the inside of SSDs: FTL and NAND-timing features merged upstream into FEMU, HyZNS, a hybrid zoned/conventional SSD I implemented across the stack, and FEMonitor, my thesis tool for telling emulator bugs apart from real storage behavior.
About
What I Build
I'm a Master's student in the Software Platforms Lab at Hanyang University. My work sits between the operating system and the SSD, where the NVMe interface, the FTL, and NAND behavior meet real workloads. Next, I want to build SSD firmware.
I've contributed nine pull requests to FEMU that are now upstream: a shared channel bus and program/erase suspend in the NAND timing model, per-open-zone ZNS write caches, zone reset with erase latency, and DSM/TRIM. I implemented HyZNS, a ZNS SSD whose conventional area can be resized online, across the stack: the emulated device, a host-side FTL in the Linux kernel, and F2FS and ZenFS support. And I built FEMonitor, which exposes FEMU's internal state to the guest. My M.S. thesis uses it to tell emulator defects and modeling artifacts apart from real storage-system behavior: it found two FEMU bugs, a channel-model artifact that inverted zone-placement results, and the causes of a 30% F2FS multi-device slowdown.
Work
Projects & Open Source
SSD internals, kernel modules, and open-source work
A ZNS SSD whose conventional area grows and shrinks online, with no reformat or data loss, by moving an area boundary inside a single namespace. Responsible for the overall implementation across the stack
CFEMULinux KernelDevice MapperNVMe ZNSF2FSZenFS
My M.S. thesis project: exposes FEMU's internal state (GC, superblocks and zones, write cache, per-channel and per-plane waits, latency histograms) through NVMe vendor-specific admin commands, at most 0.2% IOPS overhead at 1 kHz polling
CNVMeFEMUQEMUncursesPython
Designed, built, and deployed the official website for the Software Platforms Lab
Next.jsTailwind CSSPostgreSQLPrisma
Built the official website for the IEEE International Conference on Fog and Edge Computing (ICFEC) 2026
Next.jsTailwind CSS
Nine merged pull requests (2024–2026) that bring FEMU's NAND timing model and ZNS write path closer to real SSDs
- NAND timing: a shared channel-bus model for ZNS, and program/erase suspend that lets reads preempt GC (read p99 under GC from 56 ms to 0.14 ms)
- ZNS write caches: one per open zone plus fixes on the cache path, so 8 concurrent zone writers reach 275K IOPS instead of collapsing to 20K
- Zone reset with erase latency per channel/way/plane, and DSM/TRIM for the page-mapped FTL (2x fxmark throughput after deletes)
Research
Publications
Research on ZNS SSDs, FTL design, and SSD emulation
InternationalMay 2026
Lifetime-Aware Zone Allocation for ZNS SSDs
Jinyoung Kim, Doeun Kim, Junmo Seong, Hyuck Han, Sooyong Kang
IEEE International Conference on Fog and Edge Computing (ICFEC)
InternationalMay 2026
Practical Monitoring Tool for ZNS SSD Emulator
Junmo Seong, Doeun Kim, Hyuck Han, Sooyong Kang
IEEE International Conference on Fog and Edge Computing (ICFEC)
DomesticKSC 2025 Outstanding Presentation Paper AwardDec. 2025
A Real-time Monitoring Tool for Device-level Analysis in FEMU-based Storage Systems
FEMU 기반 스토리지 시스템의 디바이스 레벨 분석을 위한 실시간 모니터링 툴
Junmo Seong, Doeun Kim, Hyuck Han, Sooyong Kang
Korean Institute of Information Scientists and Engineers, KSC 2025
DomesticDec. 2024
Performance Analysis of DBSCAN Algorithms Using PIM Architecture
PIM 아키텍처를 활용한 DBSCAN 알고리즘 성능 분석
Junmo Seong, Gun Heo, Hyuck Han, Sooyong Kang
Korean Institute of Information Scientists and Engineers, KSC 2024
Background
Education & Experience
Academic background and professional journey
Education
Now
Mar. 2025 – Feb. 2027 (expected)
M.S. in Computer Science
Hanyang University
- Research Focus: FTL design, ZNS SSDs, SSD emulation, Linux storage stack
- Thesis (in progress): FEMonitor: Device-Level Observability for Validation and Diagnosis of FEMU
- Lab: Software Platforms Lab (Advisor: Prof. Sooyong Kang)
Mar. 2021 – Feb. 2025
B.S. in Computer Science
Hanyang University
- GPA: 4.1/4.5
Experience
2025 – 2026
Teaching Assistant
Hanyang University
Operating SystemsSpring 2025, Spring 2026
- Migrated the course's xv6 materials from x86 to RISC-V, aligning with MIT's current version
- Wrote assignments on MLFQ and CFS schedulers, kernel threads, copy-on-write, mmap, and file systems
- Built an automated grading system for xv6 assignments on GitHub Classroom
Capstone Design2025, 2026
- 2025: mentored a BBSSD vs. ZNS (dm-zap) performance study and dm-zap code improvements
- 2026: mentored dm-lsm, an LSM-tree device-mapper target that runs unmodified ext4/XFS on ZNS SSDs
Sep. 2023 – Feb. 2025
Research Intern
Software Platforms Lab · Hanyang University, Seoul, Korea
- Implemented and analyzed DBSCAN on the UPMEM Processing-in-Memory architecture, resulting in a published paper (KSC 2024)
- Developed multi-threaded CPU and CUDA implementations for a study of highly parallel regular-expression matching (published in IEEE Access 2025)
- Studied FEMU and the Linux kernel I/O stack (VFS, F2FS, block layer, NVMe driver) in depth, laying the groundwork for my SSD research
Expertise
Skills
Technical competencies and tools
Languages
CC++PythonTypeScript
SSD Internals
FTLGarbage CollectionNAND latency modelingP/E suspendZNS SSDTRIM / DSMOpen-Channel SSDFEMU
Host & Kernel
NVMeNVMe ZNS command setnvme-cliLinux block layerdevice mapperF2FSRocksDB / ZenFS
Benchmark
fiofxmarkdb_benchYCSB
Web
Next.jsReactTailwind CSS