Pith. sign in

REVIEW 2 major objections 5 minor 36 references

LAVA: Lifetime-Aware VM Allocation with Learned Distributions and Adaptation to Mispredictions

T0 review · 2 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Repredicting VM lifetimes beats one-shot prediction for cloud scheduling

desk verdict Credible production evidence for reprediction-based VM scheduling, but the 'fundamental advance' theorem doesn't prove what it claims; still deserves a serious review. read the letter →

arxiv 2412.09840 v2 pith:G6BHXDRM submitted 2024-12-13 cs.DC

Jianheng Ling , Pratik Worah , Yawen Wang , Yunchuan Kong , Anshul Kapoor , Chunlei Wang , Clifford Stein , Diwakar Gupta
show 11 more authors
Jason Behmer Logan A. Bush Prakash Ramanan Rajesh Kumar Thomas Chestna Yajing Liu Ying Liu Ye Zhao Kathryn S. McKinley Meeyoung Park Martin Maas
This is my paper · ORCID
classification cs.DC
keywords VMschedulinglifetimepredictionrepredictionclouddatacenterssurvivalanalysisbinpackingresourcestrandingemptyhosts
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

The paper argues that a cloud VM scheduler should not trust a single lifetime prediction made when a VM is created. Instead, it should repeatedly repredict remaining lifetimes from learned lifetime distributions, and adapt its placement decisions when earlier predictions turn out wrong. The authors claim this reprediction strategy is fundamentally more effective than one-shot prediction, backing the claim with a proof that a best-fit scheduler without reprediction needs $\Omega(m)$ more hosts than the same scheduler with reprediction when initial prediction errors are a positive constant, and with a fleet-wide production deployment that increased empty hosts by 2.3-9.2 percentage points, reduced CPU and memory stranding by about 3% and 2%, and cut VM migrations by 4.5%.

What carries the argument

The load-bearing mechanism is the lifetime distribution model combined with reprediction. A gradient-boosted decision tree is trained to output the conditional expected remaining lifetime given uptime, $E(T_r \mid T_u)$, which turns a regression model into a survival model by injecting uptime as a feature. At scheduling time and during each VM's life, the scheduler draws updated exit times from this model and detects mispredictions such as hosts that should have emptied but have not. NILAS quantizes the temporal cost $\Delta T = \max(\text{predicted VM exit} - \text{host exit}, 0)$ into buckets as a tie-breaker in the scoring function; LAVA places hosts in open or recycling states across lifetime classes (LC1--LC4) and promotes or demotes a host's class when residual VMs exit or overstay; LARS migrates longest-lived VMs first during defragmentation. The supporting theorem compares a best-fit algorithm with and without learning in a two-lifetime model and shows an $\Omega(m)$ gap in hosts required.

What would settle it

Run the paper's simulation on the same production traces with a one-shot predictor that is given every VM's true lifetime at creation time, and compare empty-host percentages to NILAS with repredictions; if the one-shot oracle matches or beats NILAS, the claimed $\Omega(m)$ advantage of reprediction would be contradicted.

Watch

Extended reading notes

Core claim

The paper's central claim is that repredicting VM and host lifetimes, and actively responding to mispredictions, delivers a fundamental advance over one-shot lifetime prediction for VM scheduling. The authors introduce LAVA (Lifetime Aware VM Allocation), which models VM lifetimes as probability distributions and uses the conditional expected remaining lifetime $E(T_r \mid T_u)$ to update each VM's exit time as it runs. Three scheduling algorithms exploit these updated lifetimes: NILAS adds a quantized temporal-cost score to an existing scheduler, LAVA deliberately places short-lived VMs on hosts with long-lived VMs so that host lifetimes are not extended, and LARS orders live migrations during defragmentation by remaining lifetime to reduce disruptions. The paper also proves that with constant positive initial prediction error, a best-fit scheduler that repredicts requires asymptotically fewer hosts, by $\Omega(m)$, than the same scheduler that does not. Production measurements and high-fidelity simulations on production traces are presented as confirmation that the theoretical advantage appears in practice.

Load-bearing premise

The proof assumes that after a job has run for a short fixed time $S$, the scheduler learns the job's true lifetime class, but in production the scheduler only receives a noisy predicted remaining lifetime, so the proof's clean 'learning after $S$' step is stronger than what the deployed system actually provides.

Editorial extensions

If this is right

  • Reprediction makes lifetime-aware scheduling robust to imperfect predictions, so the approach remains effective when the ML model degrades or encounters unfamiliar workloads.
  • More empty hosts directly increase the capacity available for large VMs, speed up maintenance rollouts, and reduce idle power consumption.
  • Reducing CPU and memory stranding recovers roughly 2-3% of cluster capacity, which the paper equates to about 1% of a cluster's capacity per percentage point.
  • LARS's lifetime-ordered migration reduces live migrations by about 4.5%, cutting disruption during defragmentation and maintenance.
  • With oracle lifetimes and ideal placement, NILAS approaches the theoretical maximum of empty hosts, indicating the greedy reprediction policy is close to optimal.

Reading between the lines

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

  • The theorem's idealized assumptions (two lifetime classes, constant arrival rate, ground truth revealed after $S$ time units) likely bound the real-world advance; production gains are finite and vary across pools, as the paper itself reports occasional inversions.
  • The reprediction principle transfers beyond VM scheduling to any allocation problem with heterogeneous object lifetimes, such as container orchestration, storage tiering, or resource management in serverless computing.
  • The paper's comparison uses a two-hour cutoff for the LA baseline; an evaluation against the production-deployed DPBFR variant of LA might show a smaller gap, since DPBFR is already designed for misprediction tolerance.
  • Because the production model only supplies a noisy conditional expectation, the asymptotic $\Omega(m)$ advantage should be tested empirically across pools with different prediction accuracies rather than taken as a guaranteed rate.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper presents LAVA, a lifetime-aware VM scheduling system for Google's Borg infrastructure. It predicts VM lifetime distributions, repredicts remaining lifetimes conditional on uptime, and uses three scheduling algorithms (NILAS, LAVA, and LARS) to improve bin packing, reduce resource stranding, and reduce live migrations. Production A/B tests and whole-pool CausalImpact analyses report empty-host increases of 2.3-9.2 percentage points, CPU stranding reductions of about 3%, memory stranding reductions of about 2%, and a 4.5% reduction in VM migrations; simulation studies on 24 C2 pools compare favorably with the LA-Binary baseline. The paper also claims a theoretical result (Theorem 1, Appendix E) that repredictions give an Omega(m) asymptotic advantage over one-shot prediction in the number of hosts required. That theoretical claim is the focus of the major concerns below.

Significance. The production-scale evaluation is the main strength of this paper: the A/B pilots, whole-pool CausalImpact analyses, and a simulator validated against production traces provide credible evidence that NILAS improves empty hosts and reduces stranding in a real hyperscale cloud. The comparison with a faithful implementation of LA-Binary in simulation, including oracle-lifetime runs, is a useful contribution. However, the theoretical 'fundamental advance' claim is not currently supported: Theorem 1's proof contains a reversed inequality and relies on an oracle-revelation assumption that the deployed conditional-expectation model does not satisfy. The practical contribution is significant, but the theoretical component needs major revision or a carefully limited restatement.

major comments (2)
  1. [Appendix E, Theorem 1 and Eq. (1)] The proof of Theorem 1 does not establish the claimed lower bound. With N = rho*lambda*x, the event probability is Pr[V] = 1 - (1 - epsilon)^N, which is an increasing function of N. The condition x <= 1/(epsilon*rho*lambda) gives N <= 1/epsilon, so Pr[V] <= 1 - (1 - epsilon)^(1/epsilon), not the displayed '>='. The supporting inequality (1 - 1/z)^z >= 1/e is also false for finite z > 1; the correct direction is '<='. The later steps ('we will need to have cm misclassified jobs', 'we can now tolerate cm errors', and the Chernoff application) are asserted without derivations or parameter specification. Since Section 4.1 and the abstract's 'fundamental advance' claim rest on this theorem, the asymptotic separation must either be proven correctly or the claim must be substantially weakened.
  2. [Appendix E; Section 3] Even if the probabilistic argument in Appendix E were repaired, Theorem 1 models a setting in which, after a job has run for S units, the scheduler learns the true short/long class. The production model of Section 3 only predicts E(Tr | Tu), a conditional expected remaining lifetime that retains prediction uncertainty; it never reveals the ground-truth class. Thus Theorem 1 is not a model of LAVA's deployed reprediction mechanism. The observed production and simulation gains may still be valid, but they are not consequences of this theorem. The paper should either prove the separation for noisy conditional expectations or explicitly limit the theoretical claim to the simplified oracle-revelation model.
minor comments (5)
  1. [Abstract; Section 7] The abstract states that 'our approach runs in production,' but Section 7 reports fleet-wide deployment only for NILAS; LAVA and LARS are validated in simulation. Please clarify the production scope to avoid overclaiming deployment of all three algorithms.
  2. [Section 4.2 and Figure 4] The 'host exit time' is defined as the maximum of the repredicted remaining VM lifetimes on the host, but Figure 4 labels it as an absolute exit time (e.g., 10h). Please clarify the convention (current time plus remaining lifetime) so that temporal cost computations are unambiguous.
  3. [Section 3] The phrase '12.5%, 25%, ... of the original lifetime' should specify the exact set of uptime fractions used to augment training examples, since this determines the training distribution and the behavior of the conditional model.
  4. [Table 2] The first column 'Migrations Scheduled' is not used in the comparison, and the reduction percentage is relative to the Baseline column. Please annotate the table so the comparison is unambiguous.
  5. [Appendix D, Figure 13] The caption says the figure shows 'relative improvements from LA-Binary,' but the axes are not labeled. Please add axis labels and explain what the correlation plot displays.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the headline results are empirically benchmarked; only a minor self-referential lineage from LLAMA appears and it is not load-bearing.

full rationale

LAVA's central claims rest on production A/B experiments (Table 1), CausalImpact time-series analysis, high-fidelity simulations on production traces, and an apples-to-apples comparison against a faithful LA-Binary baseline that reuses the same ML model without repredictions. The model outputs a conditional remaining-lifetime estimate E(Tr|Tu), but the reported empty-host, stranding, and migration improvements are measured against a production baseline and are not equal to the model output by construction. The Section 4.1 / Appendix E theorem assumes that the true short/long class is revealed after S time units; this is a simplifying idealization that limits how directly the theorem applies to the production E(Tr|Tu) mechanism, but it is a validity/correctness concern rather than a circular reduction. The paper cites the authors' prior LLAMA work for the on-the-fly misprediction-adjustment idea, but that self-citation is inspirational, not load-bearing: the present contribution is independently proved in the simplified model, simulated on production traces, and measured in production. No uniqueness theorem is imported from the authors, no fitted parameter is renamed as a prediction, and no known result is merely relabeled. The central derivation is therefore self-contained against external benchmarks, and any circularity is at most a minor and non-load-bearing self-referential lineage.

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

The central claim rests on a simplified theoretical model with idealized error revelation, and on production measurement assumptions. The free parameters are mostly scheduler design choices rather than fitted constants, but they do shape the reported improvements.

free parameters (6)
  • lifetime class boundaries = <1h, 1-10h, 10-100h, 100-1000h
    Hand-chosen decade boundaries for LAVA host/VM lifetime classes; algorithm behavior depends on these.
  • temporal cost buckets = {0m, 30m, 60m, 90m, 2h, 3h, 4h, 6h, 12h, 24h, 168h}
    Hand-chosen quantization for NILAS scoring; affects tie-breaking granularity.
  • misprediction threshold = 10x
    LAVA treats predictions as mispredicted when a VM exceeds its class by more than an order of magnitude.
  • recycling transition threshold = 90%
    Host moves from open to recycling when 90% of CPU or memory is occupied.
  • model lifetime cap = 168 hours (7 days)
    VM lifetimes are capped at 7 days in training to avoid a few very long VMs dominating the model.
  • migration busy time in simulation = 20 minutes
    Conservative simulation assumption for live migration cost on source and destination hosts.
assumptions (5)
  • ad hoc to paper After a job runs for S units of time, the scheduler learns whether it is short or long (ground truth revealed).
    Appendix E uses this to prove the Omega(m) separation. Production only provides E(Tr|Tu), a conditional expectation, not ground truth.
  • domain assumption VM lifetimes follow a generational distribution: most VMs are short-lived, but most resources are consumed by long-lived VMs.
    Figure 1 supports this in Google's fleet; it motivates the focus on lifetime-aware placement.
  • ad hoc to paper Jobs arrive at a constant rate lambda and each host holds at most k jobs in the theoretical model.
    Used in Theorem 1 (Appendix E); simplifies the analysis but does not match bursty production arrivals.
  • domain assumption CausalImpact estimates the causal effect of NILAS in whole-pool pilots without unobserved confounders.
    Section 6.2 uses Bayesian structural time series; this assumes no exogenous shocks coincident with rollout affect empty hosts.
  • domain assumption The lexicographic scoring function in Borg ensures that adding a temporal cost below other business metrics does not degrade higher-ranked objectives.
    Section 4.2 states the temporal cost is placed one level above bin packing and only acts as a tie breaker.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LAVA: Lifetime-Aware VM Allocation with Learned Distributions and Adaptation to Mispredictions." pith.science (2026). https://pith.science/paper/G6BHXDRM

@misc{pith2026241209840,
  author       = {Pith},
  title        = {Pith review of: LAVA: Lifetime-Aware VM Allocation with Learned Distributions and Adaptation to Mispredictions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G6BHXDRM}},
  note         = {Machine review of arXiv:2412.09840}
}
read the original abstract

Scheduling virtual machines (VMs) on hosts in cloud data centers dictates efficiency and is an NP-hard problem with incomplete information. Prior work improved VM scheduling with predicted VM lifetimes. Our work further improves lifetime-aware scheduling using repredictions with lifetime distributions versus one-shot prediction. Our approach repredicts and adjusts VM and host lifetimes when incorrect predictions emerge. We also present novel approaches for defragmentation and regular system maintenance, which are essential to our data center reliability and optimizations, and are not explored in prior work. We show repredictions deliver a fundamental advance in effectiveness over one-shot prediction. We call our novel combination of distribution-based lifetime predictions and scheduling algorithms Lifetime Aware VM Allocation (LAVA). LAVA reduces resource stranding and increases the number of empty hosts, which are critical for large VM scheduling, cloud system updates, and reducing dynamic energy consumption. Our approach runs in production within Google's hyperscale cloud data centers, where it improves efficiency by decreasing stranded compute and memory resources by ~3% and ~2% respectively. It increases empty hosts by 2.3-9.2 pp in production, reducing dynamic energy consumption, and increasing availability for large VMs and cloud system updates. We also show a reduction in VM migrations for host defragmentation and maintenance. In addition to our fleet-wide production deployment, we perform simulation studies to characterize the design space and show that our algorithm significantly outperforms the prior state of the art lifetime-based scheduling approach.

Figures

Figures reproduced from arXiv: 2412.09840 by the authors.

Figure 1
Figure 1. Distribution of VM lifetimes of scheduled VMs vs. their resource consumption. • We incorporate lifetime-based scheduling in a complex environment that features dynamic resource manage￾ment based on VM usage, VM live migrations, and hierarchical scheduling. We show reductions in re￾source stranding and VM migrations, in addition to improved bin packing quality. Similar to Maas et al. (Maas et al., 2020), we find that… view at source ↗
Figure 2
Figure 2. Lifetime distribution (PDF) of VM lifetimes. When the VM is scheduled, the expected (average) lifetime is 0.2 days. After it has run for 1 day, the expected remaining lifetime is 4 days. After 7 days, the expected remaining lifetime is 10 days. VM lifetimes. The best a traditional model could do would thus be to predict the expected (or average) lifetime for each VM [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overview of our VM allocation setup. 2.4 Baseline: Lifetime Alignment (LA) We compare against LA (the Lifetime Alignment algorithm) from Barbalho et al. (Barbalho et al., 2023). While other work on lifetime-based VM scheduling predates it, LA su￾persedes this work and, to our knowledge, is the only pub￾lished approach deployed in production data centers. At a high level, our setup ( [PITH_FULL_IMAGE:figures/full_fi… view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Overview of the NILAS algorithm. repredictions by Ω(m), where m is the number of hosts. Appendix E includes the precise theorem and proof. 4.2 NILAS: Non-Invasive Lifetime-Aware Scheduling The goal of NILAS ( [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Overview the LAVA Algorithm. of the next-shorter lifetime class. We therefore re-classify the host as one lifetime class lower (Figure 5b), classify all remaining VMs as the new residual VMs, and fill emerging gaps with even shorter-lived VMs. This process repeats unti…
Figure 6
Figure 6. Figure 6: Empty host improvements comparing NILAS, LAVA with LA-Binary, to our production baseline for 24 C2 pools in simulation. On average, LAVA increases empty hosts by 6.5 pp, compared to 6.1 pp for NILAS, and 5.0 pp for LA-Binary. hosts and report changes relative to the pr…
Figure 8
Figure 8. Figure 8: Histogram of model execution latencies [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: illustrates the relationship between varying amount of uptime and model accuracy. We observe that without using reprediction (no uptime/0th quantile), the model only has a F1 score of 0.8. The F1 score quickly rises above 0.9 after the 8th quantile (i.e., once the VM’s…
Figure 10
Figure 10. Figure 10: The model remains high accuracy weeks after training, making its maintenance overhead low for production deployment. our VM lifetime models and algorithms in simulation. Most time was spent on putting the approach in production. Explainable models One key requirement …
Figure 11
Figure 11. Figure 11: Impact of different features of our model on the predic￾tion accuracy, based on the split score of each feature [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 12
Figure 12. Figure 12: Histogram of the error of our GBDT model within the Log10 domain, for the first 10 million unique predictions of run￾ning NILAS with a trace. causes difficulties for regression models. We thus use Log10 for all lifetimes in our final model, including VM uptime. For ou…
Figure 13
Figure 13. Figure 13: Comparison of different metrics from simulations done on one pool, showing relative improvements from LA-Binary. To capture the benefit of correcting mispredictions, we con￾sider a simplified model in which there are only two job lifetimes, which we call short (S) and…
Figure 15
Figure 15. Figure 15: Performance at different levels of prediction accuracy [PITH_FULL_IMAGE:figures/full_fig_p018_15.png]
Figure 16
Figure 16. Figure 16: NILAS using oracular lifetime run at ideal setting (cold start and highest priority) achieves near-optimal performance. NI￾LAS also consistently outperforms the version of NILAS that does not use repredication or uptime. The reported numbers are aver￾aged across runni…
Figure 17
Figure 17. Figure 17: Effect of caching predictions, across 22 pools (in simu￾lation). Note that repredictions are still performed when a VM is added or removed from a host, irrespective of the refresh interval. Since all server host hardware is the same within each pool we tested, we can …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 16 canonical work pages

  1. [1]

    R., Dion, D., Moscibroda, T., and Menache, I

    Barbalho, H., Kovaleski, P., Li, B., Marshall, L., Molinaro, M., Pan, A., Cortez, E., Leao, M., Patwari, H., Tang, Z., Santos, T., Gonçalves, L. R., Dion, D., Moscibroda, T., and Menache, I. Virtual machine allocation with lifetime predictions. In MLSys, June 2023. URL https://www.microsoft.com/en-us/research/publication/virtual-machine-allocation-with-li...

    work page 2023

  2. [2]

    H., Gallusser, F., Koehler, J., Remy, N., and Scott, S

    Brodersen, K. H., Gallusser, F., Koehler, J., Remy, N., and Scott, S. L. Inferring causal impact using bayesian structural time-series models. The Annals of Applied Statistics, 9 0 (1): 0 247--274, 2015. ISSN 19326157. URL http://www.jstor.org/stable/24522418

    work page arXiv 2015

  3. [3]

    Buchbinder, N., Fairstein, Y., Mellou, K., Menache, I., and Naor, J. S. Online virtual machine allocation with lifetime and load predictions. In Abstract Proceedings of the 2021 ACM SIGMETRICS / International Conference on Measurement and Modeling of Computer Systems, SIGMETRICS '21, pp.\ 9–10, New York, NY, USA, 2021. Association for Computing Machinery....

    work page arXiv 2021

  4. [4]

    and Guestrin, C

    Chen, T. and Guestrin, C. XGBoost : A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD '16, pp.\ 785--794, New York, NY, USA, 2016. ACM. ISBN 978-1-4503-4232-2. doi:10.1145/2939672.2939785. URL http://doi.acm.org/10.1145/2939672.2939785

    arXiv 2016

  5. [5]

    Chollet, F. et al. Keras. https://keras.io, 2015

    2015

  6. [6]

    Resource central: Understanding and predicting workloads for improved resource management in large cloud platforms

    Cortez, E., Bonde, A., Muzio, A., Russinovich, M., Fontoura, M., and Bianchini, R. Resource central: Understanding and predicting workloads for improved resource management in large cloud platforms. In Proceedings of the International Symposium on Operating Systems Principles (SOSP), October 2017. URL https://www.microsoft.com/en-us/research/publication/r...

    work page 2017

  7. [7]

    Cox, D. R. Regression Models and Life-Tables . Journal of the Royal Statistical Society: Series B (Methodological), 34 0 (2): 0 187--202, 12 2018. ISSN 0035-9246. doi:10.1111/j.2517-6161.1972.tb00899.x. URL https://doi.org/10.1111/j.2517-6161.1972.tb00899.x

    arXiv 2018

  8. [8]

    lifelines: survival analysis in python

    Davidson-Pilon, C. lifelines: survival analysis in python. Journal of Open Source Software, 4 0 (40): 0 1317, 2019. doi:10.21105/joss.01317. URL https://doi.org/10.21105/joss.01317

    doi:10.21105/joss.01317 2019

Show all 36 references
  1. [9]

    and Kozyrakis, C

    Delimitrou, C. and Kozyrakis, C. Paragon: Qos-aware scheduling for heterogeneous datacenters. In Proceedings of the Eighteenth International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS '13, pp.\ 77–88, New York, NY, USA, 2013. As...

    2013

  2. [10]

    and Kozyrakis, C

    Delimitrou, C. and Kozyrakis, C. Quasar: resource-efficient and qos-aware cluster management. In Proceedings of the 19th International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS '14, pp.\ 127–144, New York, NY, USA, 2014 a . Ass...

    2014

  3. [11]

    and Kozyrakis, C

    Delimitrou, C. and Kozyrakis, C. Quasar: resource-efficient and qos-aware cluster management. SIGPLAN Not., 49 0 (4): 0 127–144, feb 2014 b . ISSN 0362-1340. doi:10.1145/2644865.2541941. URL https://doi.org/10.1145/2644865.2541941

    2014

  4. [12]

    D., Bodik, P., Kandula, S., Boutin, E., and Fonseca, R

    Ferguson, A. D., Bodik, P., Kandula, S., Boutin, E., and Fonseca, R. Jockey: guaranteed job latency in data parallel clusters. In Proceedings of the 7th ACM European Conference on Computer Systems, EuroSys '12, pp.\ 99–112, New York, NY, USA, 2012. Association for Computing Ma...

    2012

  5. [13]

    Gog, I., Schwarzkopf, M., Gleave, A., Watson, R. N. M., and Hand, S. Firmament: Fast, centralized cluster scheduling at scale. In 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI 16), pp.\ 99--115, Savannah, GA, November 2016. USENIX Association. ISBN...

    2016

  6. [14]

    Grandl, R., Kandula, S., Rao, S., Akella, A., and Kulkarni, J. J. Graphene: Packing and dependency-aware scheduling for data-parallel clusters. In Operating Systems Design and Implementation. USENIX, October 2016. URL https://www.microsoft.com/en-us/research/publication/graphe...

    2016

  7. [15]

    Yggdrasil decision forests: A fast and extensible decision forests library

    Guillame-Bert, M., Bruch, S., Stotz, R., and Pfeifer, J. Yggdrasil decision forests: A fast and extensible decision forests library. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD '23, pp.\ 4068–4077, New York, NY, USA, 2023. Assoc...

    2023

  8. [16]

    D., Katz, R., Shenker, S., and Stoica, I

    Hindman, B., Konwinski, A., Zaharia, M., Ghodsi, A., Joseph, A. D., Katz, R., Shenker, S., and Stoica, I. Mesos: A platform for Fine-Grained resource sharing in the data center. In 8th USENIX Symposium on Networked Systems Design and Implementation (NSDI 11), Boston, MA, March...

    2011

  9. [17]

    Jayaram Subramanya, S., Arfeen, D., Lin, S., Qiao, A., Jia, Z., and Ganger, G. R. Sia: Heterogeneity-aware, goodput-optimized ml-cluster scheduling. In Proceedings of the 29th Symposium on Operating Systems Principles, SOSP '23, pp.\ 642–657, New York, NY, USA, 2023. Associati...

    2023

  10. [18]

    Kaplan, E. L. and Meier, P. Nonparametric estimation from incomplete observations. Journal of the American Statistical Association, 53 0 (282): 0 457--481, 1958. ISSN 01621459, 1537274X. URL http://www.jstor.org/stable/2281868

    1958

  11. [19]

    George: Learning to place long-lived containers in large clusters with operation constraints

    Li, S., Wang, L., Wang, W., Yu, Y., and Li, B. George: Learning to place long-lived containers in large clusters with operation constraints. In Proceedings of the ACM Symposium on Cloud Computing, SoCC '21, pp.\ 258–272, New York, NY, USA, 2021. Association for Computing Machi...

    2021

  12. [20]

    and Hewitt, C

    Lieberman, H. and Hewitt, C. A real-time garbage collector based on the lifetimes of objects. Communications of the ACM, 26 0 (6): 0 419–429, June 1983. ISSN 0001-0782. doi:10.1145/358141.358147. URL https://doi.org/10.1145/358141.358147

    1983

  13. [21]

    G., Isard, M., Javanmard, M

    Maas, M., Andersen, D. G., Isard, M., Javanmard, M. M., McKinley, K. S., and Raffel, C. Learning-based memory allocation for c++ server workloads. In Proceedings of the Twenty-Fifth International Conference on Architectural Support for Programming Languages and Operating Syste...

    2020

  14. [22]

    B., Meng, Z., and Alizadeh, M

    Mao, H., Schwarzkopf, M., Venkatakrishnan, S. B., Meng, Z., and Alizadeh, M. Learning scheduling algorithms for data processing clusters. In Proceedings of the ACM Special Interest Group on Data Communication, SIGCOMM '19, pp.\ 270–288, New York, NY, USA, 2019. Association for...

    2019

  15. [23]

    W., Tumanov, A., Jiang, A., Kozuch, M

    Park, J. W., Tumanov, A., Jiang, A., Kozuch, M. A., and Ganger, G. R. 3sigma: distribution-based cluster scheduling for runtime uncertainty. In Proceedings of the Thirteenth EuroSys Conference, EuroSys '18, New York, NY, USA, 2018. Association for Computing Machinery. ISBN 978...

    2018

  16. [24]

    scikit-survival: A library for time-to-event analysis built on top of scikit-learn

    P \"o lsterl, S. scikit-survival: A library for time-to-event analysis built on top of scikit-learn. Journal of Machine Learning Research, 21 0 (212): 0 1--6, 2020. URL http://jmlr.org/papers/v21/20-729.html

    2020

  17. [25]

    VM live migration at scale

    Ruprecht, A., Jones, D., Shiraev, D., Harmon, G., Spivak, M., Krebs, M., Baker-Harvey, M., and Sanderson, T. VM live migration at scale. In Proceedings of the 14th ACM SIGPLAN/SIGOPS International Conference on Virtual Execution Environments, VEE '18, pp.\ 45–56, New York, NY,...

    2018

  18. [26]

    M., Marshall, L., Li, B., Zhou, S., Pan, A., Mellou, K., Narayanan, D., Zhu, T., Moscibroda, T., Dion, D., and Menache, I

    Sajal, S. M., Marshall, L., Li, B., Zhou, S., Pan, A., Mellou, K., Narayanan, D., Zhu, T., Moscibroda, T., Dion, D., and Menache, I. Kerveros: Efficient and scalable cloud admission control. In OSDI. USENIX, July 2023. URL https://www.microsoft.com/en-us/research/publication/k...

    2023

  19. [27]

    Omega: flexible, scalable schedulers for large compute clusters

    Schwarzkopf, M., Konwinski, A., Abd-El-Malek, M., and Wilkes, J. Omega: flexible, scalable schedulers for large compute clusters. In Proceedings of the 8th ACM European Conference on Computer Systems, EuroSys '13, pp.\ 351–364, New York, NY, USA, 2013. Association for Computin...

    2013

  20. [28]

    W., Kozuch, M

    Tumanov, A., Zhu, T., Park, J. W., Kozuch, M. A., Harchol-Balter, M., and Ganger, G. R. Tetrisched: global rescheduling with adaptive plan-ahead in dynamic heterogeneous clusters. In Proceedings of the Eleventh European Conference on Computer Systems, EuroSys '16, New York, NY...

    2016

  21. [29]

    Generation scavenging: A non-disruptive high performance storage reclamation algorithm

    Ungar, D. Generation scavenging: A non-disruptive high performance storage reclamation algorithm. In Proceedings of the First ACM SIGSOFT/SIGPLAN Software Engineering Symposium on Practical Software Development Environments, SDE 1, pp.\ 157–167, New York, NY, USA, 1984. Associ...

    1984

  22. [30]

    Verma, A., Cherkasova, L., and Campbell, R. H. Aria: automatic resource inference and allocation for mapreduce environments. In Proceedings of the 8th ACM International Conference on Autonomic Computing, ICAC '11, pp.\ 235–244, New York, NY, USA, 2011. Association for Computin...

    2011

  23. [31]

    R., Oppenheimer, D., Tune, E., and Wilkes, J

    Verma, A., Pedrosa, L., Korupolu, M. R., Oppenheimer, D., Tune, E., and Wilkes, J. Large-scale cluster management at Google with Borg . In Proceedings of the European Conference on Computer Systems (EuroSys), Bordeaux, France, 2015

    2015

  24. [32]

    Wang, P., Li, Y., and Reddy, C. K. Machine learning for survival analysis: A survey. ACM Comput. Surv., 51 0 (6), feb 2019. ISSN 0360-0300. doi:10.1145/3214306. URL https://doi.org/10.1145/3214306

    2019 doi

  25. [33]

    J., Berger, D., Kozyrakis, C., and Bianchini, R

    Wang, Y., Crankshaw, D., Yadwadkar, N. J., Berger, D., Kozyrakis, C., and Bianchini, R. Sol: safe on-node learning in cloud platforms. In Proceedings of the 27th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS '22, ...

    2022

  26. [34]

    J., Ananthanarayanan, G., and Katz, R

    Yadwadkar, N. J., Ananthanarayanan, G., and Katz, R. Wrangler: Predictable and faster jobs using fewer resources. In ACM Symposium on Cloud Computing (SoCC), November 2014. URL https://www.microsoft.com/en-us/research/publication/wrangler-predictable-faster-jobs-using-fewer-resources/

    2014

  27. [35]

    Fuxi: a fault-tolerant resource management and job scheduling system at internet scale

    Zhang, Z., Li, C., Tao, Y., Yang, R., Tang, H., and Xu, J. Fuxi: a fault-tolerant resource management and job scheduling system at internet scale. Proc. VLDB Endow., 7 0 (13): 0 1393–1404, aug 2014. ISSN 2150-8097. doi:10.14778/2733004.2733012. URL https://doi.org/10.14778/273...

    2014

  28. [36]

    and Maas, M

    Zhou, G. and Maas, M. Learning on distributed traces for data center storage systems. In Smola, A., Dimakis, A., and Stoica, I. (eds.), Proceedings of Machine Learning and Systems, volume 3, pp.\ 350--364, 2021. URL https://proceedings.mlsys.org/paper_files/paper/2021/file/efe...

    2021

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.