Pith. sign in

REVIEW 4 major objections 5 minor 46 references

Waltz: Temperature-Aware Cooperative Compression for High-Performance Compression-Based CSDs

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A temperature-aware scheduler that splits compression between the host CPU and a compression SSD prevents thermal shutdowns and boosts throughput.

desk verdict Waltz presents a plausible design and a real problem, but the no-shutdown claim is only shown in a fitted model inside FEMU, and the conclusion overstates the evidence. read the letter →

arxiv 2509.05365 v1 pith:YEYZIIHJ submitted 2025-09-04 cs.PF

Dingcui Yu , Yunpeng Song , Yiyang Huang , Yumiao Zhao , Yina Lv , Chundong Wang , Youtao Zhang , Liang Shi This is my paper · ORCID
classification cs.PF
keywords compression-basedcomputationalSSDtemperature-awareschedulingcooperativecompressionF2FSthermalthrottlingwriteamplificationsegmentcleaninglifetime
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that the overheating problem of compression-based SSDs can be solved in software rather than by adding cooling. The authors measure that device-side hardware compression raises drive temperature to emergency levels quickly, while host-side file-system compression degrades write throughput substantially. They propose Waltz, a scheduler inside the F2FS file system that moves compression and decompression jobs between the host CPU and the drive's hardware engine based on the drive's current temperature. If the scheme works as claimed, storage systems can keep compression's benefits—space savings and extended lifetime—without thermal throttling or shutdowns. The paper reports up to 391.5% higher write and 627% higher read throughput over a pure device-compression baseline, with an additional reserved-space scheme that cuts write amplification.

What carries the argument

The load-bearing mechanism is a temperature-threshold scheduler with two thresholds, Tsoft and Thard, fed by the drive's S.M.A.R.T. temperature reading once per second. Below Tsoft the device's hardware engine handles all compression and decompression; between Tsoft and Thard, decompression migrates to the host CPU; at or above Thard, compression also migrates to the host, letting the hardware engine idle so the device cools. A cooperative compression framework (CCF) reconciles the mismatched granularities and algorithms of F2FS and the CCSD so each compressed block can be decompressed by the side that knows its format, signalled through a reserved NVMe command bit. Two policies, Waltzs and

What would settle it

Run the Waltz scheduler on a real compression SSD with no in-package cooling while issuing continuous sequential writes and logging the onboard temperature sensor; if the drive reaches the thermal-emergency threshold (~86°C), enters throttling, or shuts down while following Waltz's policy, the central claim fails. A cheaper check: compare the emulator's predicted temperature trace against a real drive's sensor trace under the same workload and show a divergence large enough to change scheduling decisions.

Watch

Extended reading notes

Core claim

Waltz is a temperature-aware cooperative compression scheme that divides (de)compression work between the F2FS host file system and a compression-based computational SSD (CCSD). The paper claims that by reading the device's temperature and moving decompression to the host when the drive passes a soft threshold, then compression too when it passes a hard threshold, the scheduler prevents the thermal emergencies—throttling and in-flight shutdown—that occur when the hardware compression engine runs continuously. Because throttling is eliminated, the paper reports write throughput up to 391.5% and read throughput up to 627% higher than a throttling baseline, on average, in microbenchmarks. Two v

Load-bearing premise

The argument rests on the fitted temperature model—derived from real compression SSDs but embedded in an emulator—accurately reproducing how a real drive would heat up and cool down under Waltz's scheduling, since Waltz itself was never run on real hardware.

Editorial extensions

If this is right

  • If the central claim is right, thermal throttling and in-flight shutdown can be avoided without in-package cooling, even under continuous compressible writes that would overheat a vanilla CCSD in about eight minutes.
  • The device engine can stay the default and dominant executor—about 84% of (de)compression tasks remain on the device on average—so host CPU and memory overhead stay low while the drive cools only when close to the thresholds.
  • Write amplification and throughput become a runtime-configurable tradeoff: Waltzs reduces WAF by about 34.6% versus the baseline on real workloads, while Waltzp trades WAF for higher throughput.
  • The savings from compression can be used not just for lifetime extension but also to enlarge F2FS reserved space, which reduces segment-cleaning overhead and yields further performance gains.
  • The scheme's metadata and scheduling overheads are small (a few bytes plus a once-per-second temperature read), making it deployable inside existing F2FS without a dedicated cooling solution.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The asymmetric cost structure suggests the same two-threshold policy could govern other computational storage: whenever an accelerator's task can also run, more slowly, on the host CPU, the host can act as a thermal buffer that absorbs the expensive operation just long enough for the device to cool.
  • Because decompression is cheaper than compression on the host, one could set the soft threshold differently for read-heavy versus write-heavy workloads, or even rotate compression duties across a pool of drives to keep rack temperature down—both are testable extensions of Waltz's logic.
  • The 20% reserved-space cap is tuned offline; a natural follow-up is to make that cap track the live compression ratio and segment-cleaning frequency, letting the saved-space benefit adapt to changing workloads.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper studies thermal and performance trade-offs of host-side and device-side compression for compression-based computational SSDs (CCSDs). It reports real-device measurements showing that CCSD compression can cause thermal emergencies and shutdowns, while F2FS host-side compression causes large software-stack overhead. The authors propose Waltz, a temperature-aware cooperative compression framework that schedules (de)compression between F2FS and the CCSD based on the device temperature. Waltz includes a cooperative compression framework (CCF), a temperature-aware scheduler (TCS), and a configuration selector (TWINS) with two variants, Waltzs and Waltzp, for space- and performance-oriented optimization. It also introduces on-demand space allocation (OSA) that repurposes compression-saved space as F2FS reserved space. The evaluation uses a FEMU-based CCSD emulator with a fitted temperature model and reports large throughput improvements over a throttling baseline, as well as WAF and CPU-usage results.

Significance. If the central claims hold, the paper makes a useful contribution: it provides quantitative real-hardware evidence of the CCSD overheating problem, proposes a plausible cross-layer software-hardware scheduling mechanism, and shows how compression-saved space can be reused to improve F2FS segment cleaning. The design ideas—cooperative host/device compression and temperature-triggered offloading—are timely and likely to interest the storage systems community. However, the evaluation currently rests on an unvalidated thermal model embedded in an emulator, and the conclusion overstates the platform on which the system was evaluated. The real-device measurements and the F2FS-side implementation are strengths, but the device-side firmware behavior and the no-shutdown claim are not validated on real hardware.

major comments (4)
  1. [§6.1, §8, Fig. 12] The central claim that Waltz 'prevents thermal emergencies under all workloads' is not established by a real-device experiment. Section 6.1 states that the real CCSD implementation was 'circumvented' by first building a temperature model from real ASIC-based CCSDs and integrating the fitted model into FEMU. Fig. 12 reports that Waltzs/Waltzp temperatures are 'derived from the model in the emulator,' while only the Baseline temperature is read from the in-package sensor. Yet §8 says 'Evaluations on the real CCSD platform.' No model equation, fitted parameters, calibration residuals, or validation against held-out traces are provided. Since the scheduler's safety depends on the model reproducing real thermal dynamics (time constants, cooling, sensor lag), the no-shutdown claim is currently a simulation result. This must be addressed either by validating the fitted model against real closed
  2. [§4.5.2, Fig. 3] TCS sets Tsoft=76°C and Thard=85°C, with T_emergency=86°C, and reads the temperature once per second. Fig. 3 shows that on the real device the temperature continues to rise for an extended period even after throttling starts at 76°C. With only a 1°C margin between Thard and the emergency shutdown temperature, and with 1s sampling, the paper must demonstrate that the scheduler's reaction—reading S.M.A.R.T., deciding, and issuing host offload—cannot overshoot 86°C on the real device. The fitted emulator model may not capture this thermal inertia or the command/sensor latency. No worst-case latency analysis or sensitivity study is provided. This is load-bearing for the paper's primary claim of preventing in-flight shutdowns.
  3. [§6.1, §5.2, Fig. 11] Several key parameters are calibrated on the same offline measurements used to frame the evaluation: Tsoft and Thard are derived from Fig. 3, rsr_max is set to 20% from Fig. 11, and FTsc, FTc, Ts are chosen from an offline study described in §6.1. The paper gives no sensitivity analysis for these thresholds. This is a concern because the performance and thermal-safety results may be overfit to the specific device and workload configuration. At minimum, the authors should vary Tsoft (and preferably RSmax and the frequency thresholds) and report the effect on throughput and temperature. They should also discuss possible oscillation or hysteresis behavior of the TCS state machine, which is not addressed.
  4. [§6.3, Figs. 14–17] Headline numbers such as 'up to 391.5% and 627%' write/read throughput improvement over Baseline, and '95.5%' for Waltzs over TCS, are reported as averages over three runs, but the figures show no error bars, confidence intervals, or per-run spread. Given that the emulator's timing behavior under Waltz scheduling is validated only indirectly, the magnitude of these improvements should be treated as preliminary. The paper should include variance information and, ideally, a comparison of emulator timing against the real CCSD to support the throughput claims.
minor comments (5)
  1. [Table 2] The column header 'Tc == Thard' should be 'Tc >= Thard' (or 'Tc > Thard') to match the TCS description in §4.3, which uses 'greater than or equal to.'
  2. [§3.4] The sentence 'Since neither of these schemes do not uses CCSDs' has a grammar error and should read 'neither scheme uses CCSDs.'
  3. [§6.2, Fig. 13] The distribution of (de)compression tasks in Fig. 13 is said to be 'calculated based on the ratios of temperature rise and fall times and the throughput of the host side and device side on the CCSD product.' This indirect method is unclear; please describe exactly how the percentages were computed, or provide direct instrumentation measurements.
  4. [Artifacts] The paper does not provide an artifact link, the fitted temperature model, or the FEMU configuration scripts. Since the evaluation depends critically on the emulator model, making these available would substantially improve reproducibility and reviewer confidence.
  5. [Fig. 3] The figure caption uses 'Thru.' without defining the abbreviation; please spell out 'Throughput.' Also, the two Y axes (temperature and throughput) should be labeled more clearly to avoid confusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found: Waltz's claims rest on a fitted-plant simulation and a non-load-bearing self-citation, neither of which reduces the result to its inputs.

full rationale

The paper's derivation is self-contained in the circularity sense. The temperature model in §6.1 is fitted from real ASIC-based CCSD measurements and then used as the plant in FEMU; Waltz is a controller tested against that plant. The no-shutdown outcome is a closed-loop simulation output, not a restatement of the fitted parameters, and the fitted model is not refit to force the conclusion. The thresholds Tsoft=76°C and Thard=85°C are measured device characteristics used as control setpoints, not predictions. The self-citation of COCO [32] is used only for motivation/comparison, and no Waltz mechanism depends on it. The notable weakness is evidentiary, not circular: §8's 'Evaluations on the real CCSD platform' overstates §6.1's admission that real firmware changes were 'circumvented' and that Waltz temperatures are 'derived from the model in the emulator' while only Baseline is sensor-read; this is a validation/accuracy gap, not a definitional reduction.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claims rest on a fitted thermal model, hand-tuned thresholds, and the assumption that the emulator reproduces real CCSD behavior. No new physical entities are introduced.

free parameters (6)
  • Tsoft = 76°C
    Temperature threshold for moving decompression to host; set to the throttling-triggered temperature measured on real CCSDs (Fig. 3, Section 4.5.2).
  • Thard = 85°C
    Temperature threshold for moving compression to host; set just below the thermal emergency temperature of 86°C (Fig. 3, Section 4.5.2).
  • RSmax (rsr_max) = 20% of device capacity
    Cap on extra reserved space; chosen from offline throughput vs RS size study (Fig. 11, Section 6.1).
  • Ts = 1% of device capacity
    Unit of reserved-space expansion/shrink in OSA; set in Section 5.2.
  • FTsc = 4 times/second
    Segment cleaning invocation frequency threshold in OSA; set in Section 5.2.
  • FTc = 256 times/second
    Data copying frequency threshold in OSA; set in Section 5.2.
assumptions (5)
  • domain assumption The temperature model fitted from real ASIC-based CCSDs predicts CCSD temperature accurately under arbitrary workloads.
    Section 6.1 says the authors 'first building a temperature model from real ASIC-based CCSDs' and 'integrate the fitted temperature modeling into the emulator.' All Waltz temperature and throughput results come from this model.
  • domain assumption FEMU emulates the performance of the real CCSD closely enough for the comparisons.
    Section 6.1: collected performance characteristics under different temperature states to guide FEMU configuration.
  • domain assumption Real workloads' compression ratios follow a Zipfian distribution.
    Section 6.1: 'we assume that the compression ratio of real workloads follows the Zipfian distribution.'
  • domain assumption Host-side decompression costs less than host-side compression, so the scheduler can move decompression first.
    Observation-2 in Section 3.3, used to set Tsoft < Thard in Section 4.3.
  • domain assumption Reserved space beyond 20% of device capacity yields negligible benefit.
    Fig. 9 and Fig. 11 show diminishing returns; used to set rsr_max in Sections 5.1 and 6.1.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Waltz: Temperature-Aware Cooperative Compression for High-Performance Compression-Based CSDs." pith.science (2026). https://pith.science/paper/YEYZIIHJ

@misc{pith2026250905365,
  author       = {Pith},
  title        = {Pith review of: Waltz: Temperature-Aware Cooperative Compression for High-Performance Compression-Based CSDs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YEYZIIHJ}},
  note         = {Machine review of arXiv:2509.05365}
}
read the original abstract

Data compression is widely adopted for modern solid-state drives (SSDs) to mitigate both storage capacity and SSD lifetime issues. Researchers have proposed compression schemes at different system layers, including device-side solutions like CCSDs ( c ompression-based c omputational SSDs) and compression supported by host-side, like F2FS (flash-friendly file system). We conduct quantitative studies to understand how host-side and device-side compression schemes affect the temperature and performance of SSD-based storage systems. From our experiments, device-side compression, facilitated by a hardware compression engine, can raise the temperature of CCSDs to intolerable levels, resulting in throttling and service shutdown. In contrast, host-side compression causes software-stack overhead, which often results in large performance degradation and resource consumption. To ensure efficient data compression with high performance and better temperature control, we propose Waltz, a temperature-aware cooperative compression method that schedules (de)compression tasks at the host and device sides by monitoring device temperature. Furthermore, we introduce two variants (Waltzs and Waltzp) for space and performance optimization, respectively. Waltz is implemented within F2FS, achieving high performance while extending SSD lifetime and preventing overheating-induced in-flight shutdowns.

Figures

Figures reproduced from arXiv: 2509.05365 by the authors.

Figure 1
Figure 1. shows the architecture of CCSD. Compared with traditional SSDs, the added and modified modules are highlighted with red dashed boxes. The hardware-implemented (ASIC or FPGA) scheme compresses inputs with a fixed length, e.g., 4KB. Suppose that the user needs to write 16KB data to the storage. The CCSD receives and splits the data into four 4KB chunks and compresses each 4KB individually. In addition, the hardware-im… view at source ↗
Figure 2
Figure 2. The cluster layout of the compressed data in F2FS. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Temperature variation of CCSD with (W/, red line, Y1) and without (W/o, blue line, Y1) compression [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (16 more)
Figure 4
Figure 4. Figure 4: The throughput of F2FS with or without compression (W/o Compr.). The device side is CCSD with [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: The CPU usage of F2FS with or without compression (W/o Compr.). The device side is CCSD with [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: An overview of Waltz [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Workflow of CCF: ③, ⑤, and ⑥ refer to decompression on the read path, and ② and ④ refer to compression on the write path. , Vol. 1, No. 1, Article . Publication date: September 2025 [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Workflow of TCS. The self-monitoring, analysis, and reporting technology information (S.M.A.R.T [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: The WAF and throughput of F2FS with CCSD at different RS sizes (before throttling). [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Workflow of OSA. The arbiter adjusts the RS based on the metrics. [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: The effect of the RS size on performance (before throttling). [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 12
Figure 12. Figure 12: Temperature variation in four access patterns. Waltzs and Waltzp are designed to avoid the overheating [PITH_FULL_IMAGE:figures/full_fig_p016_12.png]
Figure 13
Figure 13. Figure 13: Distribution of (de)compression tasks on the host- and device-side. [PITH_FULL_IMAGE:figures/full_fig_p017_13.png]
Figure 14
Figure 14. Figure 14: Normalized average throughput without SC. 6.3.2 With SC [PITH_FULL_IMAGE:figures/full_fig_p018_14.png]
Figure 15
Figure 15. Figure 15: Normalized average throughput with SC. Waltzp achieves larger throughput improvements compared to the setting without SC. Since SC operations block user requests and degrade the system performance, the write throughput of F2FSC decreases by 65.5% while the read throug…
Figure 16
Figure 16. Figure 16: Normalized average throughput by varying compression ratios [PITH_FULL_IMAGE:figures/full_fig_p019_16.png]
Figure 17
Figure 17. Figure 17: Normalized average throughput with SC. , Vol. 1, No. 1, Article . Publication date: September 2025 [PITH_FULL_IMAGE:figures/full_fig_p019_17.png]
Figure 18
Figure 18. Figure 18: WAF under real-world workloads with SC. 6.4.3 CPU Usage and Memory Overhead. After demonstrating Waltz’s superior performance, we turn to evaluating Waltz’s host-side CPU usage and memory overhead. Sysbench [11] was used to evaluate CPU performance. The CPU performanc…
Figure 19
Figure 19. Figure 19: The performance of computation-intensive workloads under running both computation-intensive [PITH_FULL_IMAGE:figures/full_fig_p021_19.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 43 canonical work pages

  1. [1]

    CSD3000 Solid State Drive

    2024. CSD3000 Solid State Drive. https://scaleflux.com/products/csd-3000/

    work page 2024

  2. [2]

    Dell EMC PowerMax

    2024. Dell EMC PowerMax. https://www.dell.com/en-tw/blog/rise-of-data-centric-computing-with-computational- storage/

    work page 2024

  3. [3]

    2024. e2compr. http://e2compr.sourceforge.net/

    work page 2024

  4. [4]

    F2FS compression

    2024. F2FS compression. https://en.wikipedia.org/wiki/F2FS

    work page 2024

  5. [5]

    f2fs.git

    2024. f2fs.git. https://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs.git/

    work page 2024

  6. [6]

    FIO: Flexible I/O Tester

    2024. FIO: Flexible I/O Tester. https://github.com/axboe/fio

    work page 2024

  7. [7]

    Flashreduce data reduction in pure storage

    2024. Flashreduce data reduction in pure storage. http://www.purestorage.com/flasharray/flashreduce.html

    work page 2024

  8. [8]

    Ntfs compressed files

    2024. Ntfs compressed files. http://www.ntfs.com/ntfs-compressed.htm

    work page 2024

Show all 46 references
  1. [9]

    SAMSUNG SmartSSD

    2024. SAMSUNG SmartSSD. https://semiconductor.samsung.com/ssd/smart-ssd/

    2024

  2. [10]

    ScaleFlux Computational Storage

    2024. ScaleFlux Computational Storage. https://scaleflux.com/

    2024

  3. [11]

    Scriptable database and system performance benchmark

    2024. Scriptable database and system performance benchmark. https://github.com/akopytov/sysbench

    2024

  4. [12]

    Keivan Alizadeh, Iman Mirzadeh, Dmitry Belenko, Karen Khatamifard, Minsik Cho, Carlo C Del Mundo, Mohammad Rastegari, and Mehrdad Farajtabar. 2024. LLM in a flash: Efficient Large Language Model Inference with Limited Memory. arXiv:2312.11514 [cs.CL] https://arxiv.org/abs/2312.11514

    2024 arXiv

  5. [13]

    Ganger, and George Amvrosiadis

    Matias Bjørling, Abutalib Aghayev, Hans Holmberg, Aravind Ramesh, Damien Le Moal, Gregory R. Ganger, and George Amvrosiadis. 2021. ZNS: Avoiding the Block Interface Tax for Flash-based SSDs. InUSENIX Annual Technical Conference (USENIX ATC 21). USENIX Association, 689–703. htt...

    2021

  6. [14]

    Khadija Bouraqia, Essaid Sabir, Mohamed Sadik, and Latif Ladid. 2020. Quality of Experience for Streaming Services: Measurements, Challenges and Insights. IEEE Access 8 (2020), 13341–13361. https://doi.org/10.1109/ACCESS.2020. 2965099

    2020 doi

  7. [15]

    Xiang Gao, Mingkai Dong, Xie Miao, Wei Du, Chao Yu, and Haibo Chen. 2019. EROFS: A Compression-friendly Readonly File System for Resource-scarce Devices. In USENIX Annual Technical Conference (USENIX ATC 19) . USENIX Association, Renton, WA, 149–162. http://www.usenix.org/conf...

    2019

  8. [16]

    MW Green. 1986. Pareto Distributions

    1986

  9. [17]

    Hyunho Gwak, Yunji Kang, and Dongkun Shin. 2015. Reducing garbage collection overhead of log-structured file systems with GC journaling. In International Symposium on Consumer Electronics (ISCE) . 1–2. https://doi.org/10.1109/ ISCE.2015.7177770

    2015

  10. [18]

    Xiaokang Hu, Fuzong Wang, Weigang Li, Jian Li, and Haibing Guan. 2019. QZFS: QAT Accelerated Compression in File System for Application Agnostic and Cost Efficient Data Storage. InUSENIX Annual Technical Conference (USENIX ATC 19). USENIX Association, Renton, WA, 163–176. http...

    2019

  11. [19]

    Cheng Ji, Li-Pin Chang, Riwei Pan, Chao Wu, Congming Gao, Liang Shi, Tei-Wei Kuo, and Chun Jason Xue. 2021. Pattern-Guided File Compression with User-Experience Enhancement for Log-Structured File System on Mobile Devices. In USENIX Conference on File and Storage Technologies ...

    2021

  12. [20]

    Cheng Ji, Li-Pin Chang, Liang Shi, Congming Gao, Chao Wu, Yuangang Wang, and Chun Jason Xue. 2017. Lightweight Data Compression for Mobile Flash Storage. 16, 5s, Article 183 (sep 2017), 18 pages. https://doi.org/10.1145/3126511

    2017 doi

  13. [21]

    Jeong-Uk Kang, Heeseung Jo, Jin-Soo Kim, and Joonwon Lee. 2006. A Superblock-Based Flash Translation Layer for NAND Flash Memory. In Proceedings of the 6th ACM & IEEE International Conference on Embedded Software (Seoul, Korea) (EMSOFT ’06). Association for Computing Machinery...

    2006

  14. [22]

    Taejin Kim, Duwon Hong, Sangwook Shane Hahn, Myoungjun Chun, Sungjin Lee, Jooyoung Hwang, Jongyoul Lee, and Jihong Kim. 2019. Fully Automatic Stream Management for Multi-Streamed SSDs Using Program Contexts. In USENIX Conference on File and Storage Technologies (FAST 19) . USE...

    2019

  15. [23]

    Tomer Lange, Joseph (Seffi) Naor, and Gala Yadgar. 2022. Offline and Online Algorithms for SSD Management. SIGMETRICS Perform. Eval. Rev. 50, 1 (jul 2022), 89–90. https://doi.org/10.1145/3547353.3522630

    2022

  16. [24]

    Changman Lee, Dongho Sim, Jooyoung Hwang, and Sangyeun Cho. 2015. F2FS: A New File System for Flash Storage. In USENIX Conference on File and Storage Technologies (FAST 15) . USENIX Association, Santa Clara, CA, 273–286. https://www.usenix.org/conference/fast15/technical-sessi...

    2015

  17. [25]

    Shipman, Sarp Oral, Feiyi Wang, and Jongman Kim

    Junghee Lee, Youngjae Kim, Galen M. Shipman, Sarp Oral, Feiyi Wang, and Jongman Kim. 2011. A semi-preemptive garbage collector for solid state drives. In (IEEE ISPASS) IEEE International Symposium on Performance Analysis of Systems and Software. 12–21. https://doi.org/10.1109/...

    2011

  18. [26]

    Cangyuan Li, Ying Wang, Cheng Liu, Shengwen Liang, Huawei Li, and Xiaowei Li. 2021. GLIST: Towards In- Storage Graph Learning. In USENIX Annual Technical Conference (USENIX ATC 21) . USENIX Association, 225–238. https://www.usenix.org/conference/atc21/presentation/li-cangyuan

    2021

  19. [27]

    Huaicheng Li, Mingzhe Hao, Michael Hao Tong, Swaminathan Sundararaman, Matias Bjørling, and Haryadi S. Gunawi

  20. [28]

    Dongil Park, Seungyong Cheon, and Youjip Won. 2015. Suspend-aware Segment Cleaning in Log-structured File System. In USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 15) . USENIX Association, Santa Clara, CA. https://www.usenix.org/conference/hotstorage15/...

    2015

  21. [29]

    Hyunseung Park, Eunjae Lee, Jaeho Kim, and Sam H. Noh. 2021. Lightweight Data Lifetime Classification Using Migration Counts to Improve Performance and Lifetime of Flash-Based SSDs (APSys ’21). Association for Computing Machinery, New York, NY, USA, 25–33. https://doi.org/10.1...

    2021

  22. [30]

    Ohad Rodeh, Josef Bacik, and Chris Mason. 2013. BTRFS: The Linux B-Tree Filesystem.ACM Trans. Storage 9, 3, Article 9 (aug 2013), 32 pages. https://doi.org/10.1145/2501620.2501623

    2013

  23. [31]

    Russell Sears, Mark Callaghan, and Eric Brewer. 2008. Rose: Compressed, Log-Structured Replication. Proc. VLDB Endow. 1, 1 (aug 2008), 526–537. https://doi.org/10.14778/1453856.1453914

    2008

  24. [32]

    Yunpeng Song, Yiyang Huang, Yina Lv, Yi Zhang, and Liang Shi. 2023. When F2FS Meets Compression-Based SSD!. In Proceedings of the 15th ACM Workshop on Hot Topics in Storage and File Systems (Boston, MA, USA) (HotStorage ’23). Association for Computing Machinery, New York, NY, ...

    2023

  25. [33]

    Yunpeng Song, Yina Lv, and Liang Shi. 2023. Adaptive Differential Wearing for Read Performance Optimization on High-Density NAND Flash Memory. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (2023), 1–1. https://doi.org/10.1109/TCAD.2023.3297971

    2023

  26. [34]

    Yunpeng Song, Yina Lv, and Liang Shi. 2023. DECC: Differential ECC for Read Performance Optimization on High- Density NAND Flash Memory. In Proceedings of the 28th Asia and South Pacific Design Automation Conference (Tokyo, Japan) (ASPDAC ’23). Association for Computing Machin...

    2023

  27. [35]

    Vasily Tarasov, Erez Zadok, and Spencer Shepler. 2016. Filebench: A Flexible Framework for File System Benchmarking. login Usenix Mag. 41, 1 (2016). https://www.usenix.org/publications/login/spring2016/tarasov

    2016

  28. [36]

    Jon Tate, Christian Burns, Bosmat Tuv-El, Jorge Quintal, et al. 2018. IBM Real-time Compression in IBM SAN Volume Controller and IBM Storwize V7000 . IBM Redbooks

    2018

  29. [37]

    Chao Wu, Cheng Ji, and Chun Jason Xue. 2019. Reinforcement Learning based Background Segment Cleaning for Log-structured File System on Mobile Devices. In IEEE International Conference on Embedded Software and Systems (ICESS). 1–8. https://doi.org/10.1109/ICESS.2019.8782508

    2019

  30. [38]

    Erci Xu, Mai Zheng, Feng Qin, Yikang Xu, and Jiesheng Wu. 2019. Lessons and Actions: What We Learned from 10K SSD-Related Storage System Failures. In USENIX Annual Technical Conference (USENIX ATC 19) . USENIX Association, Renton, WA, 961–976. https://www.usenix.org/conference...

    2019

  31. [39]

    Chien, and Haryadi S

    Shiqin Yan, Huaicheng Li, Mingzhe Hao, Michael Hao Tong, Swaminathan Sundararaman, Andrew A. Chien, and Haryadi S. Gunawi. 2017. Tiny-Tail Flash: Near-Perfect Elimination of Garbage Collection Tail Latencies in NAND , Vol. 1, No. 1, Article . Publication date: September 2025. ...

    2017 doi

  32. [40]

    Jiacheng Zhang, Jiwu Shu, and Youyou Lu. 2016. ParaFS: A Log-Structured File System to Exploit the Internal Parallelism of Flash Devices. In USENIX Annual Technical Conference (USENIX ATC 16) . USENIX Association, Denver, CO, 87–100. https://www.usenix.org/conference/atc16/tec...

    2016

  33. [41]

    Xuebin Zhang, Jiangpeng Li, Hao Wang, Danni Xiong, Jerry Qu, Hyunsuk Shin, Jung Pill Kim, and Tong Zhang. 2017. Realizing Transparent OS/Apps Compression in Mobile Devices at Zero Latency Overhead. IEEE Trans. Comput. 66, 7 (2017), 1188–1199. https://doi.org/10.1109/TC.2017.2664838

    2017

  34. [42]

    Xuebin Zhang, Jiangpeng Li, Hao Wang, Kai Zhao, and Tong Zhang. 2016. Reducing Solid-State Storage Device Write Stress through Opportunistic In-place Delta Compression. In USENIX Conference on File and Storage Technologies (FAST 16). USENIX Association, Santa Clara, CA, 111–12...

    2016

  35. [43]

    Ning Zheng, Xubin Chen, Jiangpeng Li, Qi Wu, Yang Liu, Yong Peng, Fei Sun, Hao Zhong, and Tong Zhang. 2020. Re-think Data Management Software Design Upon the Arrival of Storage Hardware with Built-in Transparent Compression. In USENIX Workshop on Hot Topics in Storage and File...

    2020

  36. [44]

    You Zhou, Fei Wu, Ping Huang, Xubin He, Changsheng Xie, and Jian Zhou. 2015. An Efficient Page-Level FTL to Optimize Address Translation in Flash Memory. In Proceedings of the Tenth European Conference on Computer Systems (Bordeaux, France) (EuroSys ’15). Association for Compu...

    2015

  37. [45]

    Aviad Zuck, Sivan Toledo, Dmitry Sotnikov, and Danny Harnik. 2014. Compression and SSDs: Where and How?. In Workshop on Interactions of NVM/Flash with Operating Systems and Workloads (INFLOW 14) . USENIX Association, Broomfield, CO. https://www.usenix.org/conference/inflow14/w...

    2014

  38. [2018]

    The CASE of FEMU: Cheap, accurate, scalable and extensible flash emulator. In FAST. USENIX Association, Oakland, CA, 83–90. https://www.usenix.org/conference/fast18/presentation/li

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.