pith. sign in

arxiv: 2606.23945 · v1 · pith:APIDUIOUnew · submitted 2026-06-22 · 💻 cs.PF · cs.DC· cs.OS

LMS-AR: LMS Prediction-based Adaptive Regulator for Memory Bandwidth in Multicore Systems

Sudarshan Srinivasan , Deepak Gangadharan , Dip Goswami This is my paper

Pith reviewed 2026-06-26 05:50 UTC · model grok-4.3

classification 💻 cs.PF cs.DCcs.OS
keywords memory bandwidth regulationmulticore systemsadaptive regulatorLMS predictionmemory contentionMemguardSPEC CPU 2017Linux kernel module
0
0 comments X

The pith

LMS-AR regulates memory bandwidth from an external master core using LMS prediction to reduce contention slowdowns compared to Memguard.

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

The paper proposes LMS-AR, a kernel-module regulator that predicts per-core memory bandwidth needs with an adaptive least mean squares filter and enforces allocations from a non-dedicated master core. This external placement keeps heavy computation off the regulated cores so it does not interfere with running applications. Experiments distributing SPEC CPU 2017 benchmarks across cores show lower slowdown ratios from memory contention than the earlier Memguard approach. If the claim holds, shared memory bandwidth becomes a controllable rather than uncertain resource, raising the reliability of quality-of-service guarantees in ordinary multicore processors. The design is released as open source for direct replication.

Core claim

The central claim is that LMS-AR, an LMS prediction-based adaptive regulator running inside a Linux kernel module, can distribute shared memory bandwidth among CPU cores by monitoring and controlling from a master core that is not reserved solely for regulation, thereby producing significantly smaller slowdown ratios than Memguard when the same SPEC CPU 2017 workloads contend for bandwidth.

What carries the argument

The LMS prediction-based adaptive regulator, which forecasts each core's bandwidth demand from observed usage and adjusts throttling from outside the regulated cores.

If this is right

  • Memory bandwidth can be treated as an explicitly allocated resource rather than an unpredictable shared one.
  • Quality-of-service targets become enforceable on commodity multicore hardware without dedicated controller cores.
  • Computationally heavier prediction methods become practical because they run outside the cores whose performance must be protected.
  • Open-source kernel-module implementation allows direct measurement of the claimed slowdown improvement on other workloads.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The external-master model could be applied to regulate other contended resources such as last-level cache or interconnect bandwidth.
  • If master-core overhead stays low across varying core counts, the same structure might support dynamic allocation in latency-sensitive or real-time settings.
  • Prediction accuracy limits would be revealed by testing on workloads whose bandwidth demand changes faster than the LMS filter can track.
  • A hybrid controller combining LMS prediction with simpler threshold rules might further reduce cases where prediction error increases slowdown.

Load-bearing premise

That an external master core can monitor usage and enforce bandwidth limits accurately enough to cut slowdowns without adding unacceptable overhead or interference on the cores being regulated.

What would settle it

A side-by-side run of the same SPEC benchmark mix showing that slowdown ratios under LMS-AR equal or exceed those under Memguard, or that the master core's monitoring activity itself measurably slows the regulated cores.

Figures

Figures reproduced from arXiv: 2606.23945 by Deepak Gangadharan, Dip Goswami, Sudarshan Srinivasan.

Figure 1
Figure 1. Figure 1: Comparison of IPC performance: Unregulated vs throttled with fixed [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: System Overview: LMS-AR’s estimator-regulator components exe [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of Slowdown ratios for system with 2 cores + 1 (master core) [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Workload Performance 4 Cores + 1 master core setup with low [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Workload Performance 4 Cores + 1 master core setup with diverse [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 4
Figure 4. Figure 4: Workload Performance 4 Cores + 1 master core setup with high [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 7
Figure 7. Figure 7: Variation in IPC performance of workloads in 4 + 1 core setup with [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
read the original abstract

Memory bandwidth contention in multi-core systems severely impacts application performance and quality-of-service (QoS) guarantees. Regulating the shared memory bandwidth mitigates the memory performance uncertainty thereby making it a manageable resource and improving trustworthiness of multi-core systems. In this work we propose a memory bandwidth regulation mechanism LMS-AR, i.e., LMS Prediction-based Adaptive Regulator within a Linux kernel module to distribute the memory bandwidth as a resource among the CPU cores. We describe a design in which both monitoring and regulation is enforced from outside by a master core - which is not a dedicated controller for regulation. This allows for plugging in computationally heavy prediction and regulation algorithms without interfering with the regulated core. An adaptive filtering technique was employed for prediction of per-core bandwidth requirement. We conducted several experiments with SPEC CPU 2017 benchmarks distributed across multiple cores. Our proposed approach demonstrated significant improvement over Memguard with respect to slowdown ratios caused due to memory contention. Our solution is hosted publicly at $\href{https://github.com/ss22ongithub/LMSAdaptiveRegulator}{https://github.com/ss22ongithub/LMSAdaptiveRegulator}$.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript proposes LMS-AR, a Linux kernel module implementing an LMS (Least Mean Squares) prediction-based adaptive regulator for shared memory bandwidth in multicore systems. Monitoring and regulation are performed from an external master core (not a dedicated controller) to enable computationally intensive prediction without direct interference on regulated cores. Experiments distribute SPEC CPU 2017 benchmarks across cores and claim significant improvement in slowdown ratios relative to Memguard; the implementation is released publicly on GitHub.

Significance. If the central empirical claim is substantiated with quantitative controls, the approach could improve QoS predictability in commodity multicore platforms by showing that external adaptive filtering can allocate bandwidth without dedicated hardware. The public code release is a clear strength for reproducibility and further experimentation.

major comments (2)
  1. [Abstract] Abstract: the headline claim of 'significant improvement' in slowdown ratios over Memguard supplies no numerical values, standard deviations, workload counts, or statistical tests, so the magnitude and reliability of the result cannot be assessed from the provided text.
  2. [Abstract] Abstract (and implied experimental section): the central attribution of improvement to the LMS predictor rests on the untested assumption that the master core's own memory accesses for sampling and control add negligible contention. No ablation (master disabled vs. enabled under identical regulation policy), no separate accounting of master-core bandwidth, and no comparison of regulated-core performance with vs. without the master are described, leaving open the possibility that observed gains arise from altered experimental topology rather than the predictor.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript accordingly to strengthen the presentation of results and experimental controls.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the headline claim of 'significant improvement' in slowdown ratios over Memguard supplies no numerical values, standard deviations, workload counts, or statistical tests, so the magnitude and reliability of the result cannot be assessed from the provided text.

    Authors: We agree that the abstract should provide quantitative details to allow readers to assess the claims. In the revised manuscript we will update the abstract to report the specific average slowdown ratio improvements (with standard deviations), the exact number of SPEC CPU 2017 workloads evaluated, and any statistical tests used to establish significance. revision: yes

  2. Referee: [Abstract] Abstract (and implied experimental section): the central attribution of improvement to the LMS predictor rests on the untested assumption that the master core's own memory accesses for sampling and control add negligible contention. No ablation (master disabled vs. enabled under identical regulation policy), no separate accounting of master-core bandwidth, and no comparison of regulated-core performance with vs. without the master are described, leaving open the possibility that observed gains arise from altered experimental topology rather than the predictor.

    Authors: This is a fair point; the current manuscript does not include an explicit ablation isolating the master core's contribution. The external-master design was chosen precisely to permit computationally intensive prediction without direct interference on the regulated cores. In revision we will add measured bandwidth consumption figures for the master core and a discussion of its potential contention impact. We will also include an ablation experiment (master enabled vs. disabled under the same regulation policy) if additional runs can be completed with the existing setup; otherwise we will clearly state the limitation and the rationale for the topology. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical implementation with external validation, no derivation chain or fitted predictions.

full rationale

The paper presents an empirical Linux kernel module implementing LMS-based bandwidth regulation from an external master core, with results from SPEC CPU 2017 benchmarks showing improvement over Memguard. No equations, fitted parameters, or mathematical derivations are described in the abstract or claimed contribution. The LMS predictor is a standard adaptive filter technique applied to measured bandwidth data; the central claim rests on experimental slowdown ratios rather than any self-referential definition, renamed result, or self-citation chain. The work is therefore self-contained against external benchmarks with no load-bearing internal reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented entities; the approach implicitly relies on standard LMS filter update rules and Linux kernel primitives whose details are not stated.

pith-pipeline@v0.9.1-grok · 5732 in / 1025 out tokens · 27264 ms · 2026-06-26T05:50:23.709703+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

23 extracted references · 2 canonical work pages

  1. [1]

    Predator: a predictable sdram memory controller,

    B. Akesson, K. Goossens, and M. Ringhofer, “Predator: a predictable sdram memory controller,” inProceedings of the 5th IEEE/ACM International Conference on Hardware/Software Codesign and System Synthesis, ser. CODES+ISSS ’07. New York, NY , USA: Association for Computing Machinery, 2007, p. 251–256. [Online]. Available: https://doi.org/10.1145/1289816.1289877

    work page doi:10.1145/1289816.1289877 2007

  2. [2]

    Memguard: Memory bandwidth reservation system for efficient performance iso- lation in multi-core platforms,

    H. Yun, G. Yao, R. Pellizzoni, M. Caccamo, and L. Sha, “Memguard: Memory bandwidth reservation system for efficient performance iso- lation in multi-core platforms,” in2013 IEEE 19th Real-Time and Embedded Technology and Applications Symposium (RTAS), 2013, pp. 55–64

    2013

  3. [3]

    Mempol: polling-based microsecond-scale per-core memory bandwidth regulation,

    A. Zuepke, A. Bastoni, W. Chen, M. Caccamo, and R. Mancuso, “Mempol: polling-based microsecond-scale per-core memory bandwidth regulation,”Real-Time Systems, vol. 60, pp. 369–412, 9 2024

    2024

  4. [4]

    E-warp: A system- wide framework for memory bandwidth profiling and management,

    P. Sohal, R. Tabish, U. Drepper, and R. Mancuso, “E-warp: A system- wide framework for memory bandwidth profiling and management,” in 2020 IEEE Real-Time Systems Symposium (RTSS), 2020, pp. 345–357

    2020

  5. [5]

    Learning algorithms for adaptive processing and control,

    B. Widrow, M. A. Lehr, F. Beaufays, E. A. Wan, and M. Bilello, “Learning algorithms for adaptive processing and control,”IEEE International Conference on Neural Networks, pp. 1–8 vol.1, 1993. [Online]. Available: https://api.semanticscholar.org/CorpusID:17246524

    1993

  6. [6]

    (2026) Floating point api

    The Linux Kernel Project. (2026) Floating point api. [Online]; Accessed: March 19, 2026. [Online]. Available: https://docs.kernel.org/ core-api/floating-point.html

    2026

  7. [7]

    Intel intel® 64 and ia-32 architectures software developer’s manual

    Intel, “Intel intel® 64 and ia-32 architectures software developer’s manual.” [Online]. Available: https://www.intel.com/content/dam/ develop/external/us/en/documents/335592-sdm-vol-4.pdf

  8. [8]

    Shielding linux resources,

    openSUSE Project, “Shielding linux resources,” [Online]; Accessed: May 19, 2026. [Online]. Available: https://documentation.suse.com/ sle-rt/15-SP6/html/SLE-RT-all/book-shielding.html

    2026

  9. [9]

    Srinivasan and D

    S. Srinivasan and D. Gangadharan. Technical report on lms- ar memory bandwidth regulator. [Online]; Accessed: May 19, 2026. [Online]. Available: https://github.com/ss22ongithub/ LMSAdaptiveRegulator/blob/main/LMSARTechnicalReport.pdf

    2026

  10. [10]

    A survey of tech- niques for reducing interference in real-time applications on multicore platforms,

    T. Lugo, S. Lozano, J. Fernandez, and J. Carretero, “A survey of tech- niques for reducing interference in real-time applications on multicore platforms,”IEEE Access, vol. 10, pp. 21 853–21 882, 2022

    2022

  11. [11]

    Profiling and controlling i/o-related memory contention in cots heterogeneous platforms,

    M. Zini, G. Cicero, D. Casini, and A. Biondi, “Profiling and controlling i/o-related memory contention in cots heterogeneous platforms,”Soft- ware: Practice and Experience, vol. 52, no. 5, pp. 1095–1113, 2022

    2022

  12. [12]

    Bru: Bandwidth regulation unit for real-time multicore processors,

    F. Farshchi, Q. Huang, and H. Yun, “Bru: Bandwidth regulation unit for real-time multicore processors,” in2020 IEEE Real-Time and Embedded Technology and Applications Symposium (RTAS), 2020, pp. 364–375

    2020

  13. [13]

    Coherence-aided memory bandwidth regula- tion,

    I. Izhbirdeev, D. Hoornaert, W. Chen, A. Zuepke, Y . Hammad, M. Cac- camo, and R. Mancuso, “Coherence-aided memory bandwidth regula- tion,” in2024 IEEE Real-Time Systems Symposium (RTSS), 2024, pp. 322–335

    2024

  14. [14]

    Duomc: Tight dram latency bounds with shared banks and near-cots performance,

    R. Mirosanlou, M. Hassan, and R. Pellizzoni, “Duomc: Tight dram latency bounds with shared banks and near-cots performance,” in Proceedings of the International Symposium on Memory Systems, ser. MEMSYS ’21. New York, NY , USA: Association for Computing Machinery, 2022

    2022

  15. [15]

    Shedding the shackles of time-division multiplexing,

    F. Hebbache, M. Jan, F. Brandner, and L. Pautet, “Shedding the shackles of time-division multiplexing,” in2018 IEEE Real-Time Systems Sym- posium (RTSS), 2018, pp. 456–468

    2018

  16. [16]

    Drambulism: Balancing performance and predictability through dynamic pipelining,

    R. Mirosanlou, M. Hassan, and R. Pellizzoni, “Drambulism: Balancing performance and predictability through dynamic pipelining,” in2020 IEEE Real-Time and Embedded Technology and Applications Sympo- sium (RTAS), 2020, pp. 82–94

    2020

  17. [17]

    Architecture and analysis of a dynamically-scheduled real-time memory controller,

    Y . Li, B. Akesson, and K. Goossens, “Architecture and analysis of a dynamically-scheduled real-time memory controller,”Real-Time Systems, vol. 52, no. 5, pp. 675–729, 2016. [Online]. Available: https://doi.org/10.1007/s11241-015-9235-y

    work page doi:10.1007/s11241-015-9235-y 2016

  18. [18]

    A predictable and command-level priority-based dram controller for mixed-criticality systems,

    H. Kim, D. Broman, E. A. Lee, M. Zimmer, A. Shrivastava, and J. Oh, “A predictable and command-level priority-based dram controller for mixed-criticality systems,” in21st IEEE Real-Time and Embedded Technology and Applications Symposium, 2015, pp. 317–326

    2015

  19. [19]

    Analyzing arm’s mpam from the perspective of time predictability,

    M. Zini, D. Casini, and A. Biondi, “Analyzing arm’s mpam from the perspective of time predictability,”IEEE Transactions on Computers, vol. 72, no. 1, pp. 168–182, 2023

    2023

  20. [20]

    Memory bandwidth management for efficient performance isolation in multi-core platforms,

    H. Yun, G. Yao, R. Pellizzoni, M. Caccamo, and L. Sha, “Memory bandwidth management for efficient performance isolation in multi-core platforms,”IEEE Transactions on Computers, vol. 65, no. 2, pp. 562– 576, 2016

    2016

  21. [21]

    Mixed- criticality scheduling with dynamic memory bandwidth regulation,

    M. A. Awan, K. Bletsas, P. F. Souto, B. Akesson, and E. Tovar, “Mixed- criticality scheduling with dynamic memory bandwidth regulation,” in 2018 IEEE 24th International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA), 2018, pp. 111–117

    2018

  22. [22]

    Schedulability analysis for memory bandwidth regulated multicore real- time systems,

    G. Yao, H. Yun, Z. P. Wu, R. Pellizzoni, M. Caccamo, and L. Sha, “Schedulability analysis for memory bandwidth regulated multicore real- time systems,”IEEE Transactions on Computers, vol. 65, no. 2, pp. 601–614, 2016

    2016

  23. [23]

    Memory utilization-based dynamic bandwidth regu- lation for temporal isolation in multi-cores,

    A. Saeed, D. Dasari, D. Ziegenbein, V . Rajasekaran, F. Rehm, M. Pressler, A. Hamann, D. Mueller-Gritschneder, A. Gerstlauer, and U. Schlichtmann, “Memory utilization-based dynamic bandwidth regu- lation for temporal isolation in multi-cores,” in2022 IEEE 28th Real- Time and Embedded Technology and Applications Symposium (RTAS), 2022

    2022