Pith. sign in

REVIEW 4 major objections 4 minor 29 references

Sustainable Carbon-Aware and Water-Efficient LLM Scheduling in Geo-Distributed Cloud Datacenters

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a new scheduling framework, SLIT, co-optimizes time-to-first-token, carbon emissions, water use, and energy cost for LLM inference across geo-distributed datacenters, and in simulation reduces each by 73-99% versus…

desk verdict A genuinely new four-objective LLM scheduling framework whose headline water savings rest on a dimensionally wrong model and an unvalidated simulator; worth revising but not citing yet. read the letter →

arxiv 2505.23554 v1 pith:RO6XLY7H submitted 2025-05-29 cs.DC cs.AIcs.LG

Hayden Moore , Sirui Qi , Ninad Hogade , Dejan Milojicic , Cullen Bash , Sudeep Pasricha This is my paper
classification cs.DCcs.AIcs.LG
keywords carbon-awareschedulingwater-efficientdatacentersLLMinferencegeo-distributedcloudsmulti-objectiveoptimizationParetofrontevolutionaryalgorithmgradientboosting
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 environmental cost of serving LLM inference requests can be cut substantially by smart request scheduling across geo-distributed cloud datacenters. It introduces SLIT, a framework that co-optimizes four objectives at once: time-to-first-token (TTFT), carbon emissions, water usage, and energy cost, returning a Pareto set of scheduling plans rather than a single answer. In simulation, the single-objective SLIT variants reduce carbon emissions by 98-99%, water usage by 97-99%, energy cost by 96-99%, and TTFT by 73-81% compared with the Helix and Splitwise baselines. A balanced variant, SLIT-Balance, beats Helix on all four metrics and Splitwise on three while keeping TTFT competitive. If the underlying datacenter models are faithful, the paper's approach would let LLM providers shrink the inference-phase footprint that now dominates LLM costs.

What carries the argument

The mechanism that carries the argument is the SLIT metaheuristic, iterating ML-guided local search with an evolutionary algorithm. The search begins from a partially random population seeded with two extreme plans (even distribution and single-location scheduling); each plan is locally searched, and the trajectories are used to retrain a gradient-boosting predictor every $freq$ iterations. An evolutionary loop then randomly selects parent plans from the current nondominated population, applies crossover and mutation to generate children, and keeps only dominant plans in the population, yielding a real-time Pareto front. Every plan is scored against the paper's datacenter model: IT energy from power states (Eq. 5), cooling energy with $E_{cooling}=3\times E_{CRAC}$ (Eq. 8), a 13% power-conditioning overhead (Eq. 9), water from evaporative, blowdown, and grid components (Eqs. 12-15), and carbon from grid intensity plus water-treatment emissions (Eqs. 16-18). The workload predictor (a set of incrementally trained linear regression models) supplies the request count for the upcoming 15-minute epoch.

What would settle it

Run SLIT's recommended plans on a small real geo-distributed testbed for 24 hours, recording actual energy, water, carbon, and TTFT, and compare with simulator predictions; large discrepancies in cooling or water overhead would show the reductions are simulation artifacts.

Watch

Extended reading notes

Core claim

The central claim is that scheduling LLM inference requests is the key lever for sustainability: by choosing which of several global datacenters (with heterogeneous A100/H100 nodes, different grid carbon intensities, water intensities, time-of-use electricity prices, and cooling costs) handles each request, a provider can simultaneously cut carbon, water, and energy cost without sacrificing perceived latency. SLIT finds these plans by combining an ML-guided local search, in which a gradient-boosting model trained on previous search trajectories suggests promising starting points, with an evolutionary algorithm that crosses and mutates existing good plans to escape local optima. Over a 24-hour scaled workload derived from the BurstGPT trace, the paper reports that SLIT's Pareto-optimal solutions dominate the Helix mixed-integer-programming scheduler and the Splitwise phase-splitting queue scheduler on the co-optimization objectives, with per-epoch plots showing the balanced SLIT solution staying near Splitwise's TTFT while emitting far less carbon and using far less water.

Load-bearing premise

The load-bearing premise is that the paper's equations for datacenter energy, cooling, water, and carbon (with cooling set to three times CRAC energy and a 13% power-conditioning overhead) faithfully describe how real datacenters behave, since the simulator is never checked against measured data.

Editorial extensions

If this is right

  • Cloud operators can select, from the Pareto front SLIT returns, a plan that minimizes any single objective (carbon, water, cost, or TTFT) or a balanced compromise, within the 15-minute decision window.
  • SLIT-Balance outperforms Helix on all four objectives and Splitwise on carbon, water, and cost while keeping TTFT competitive, showing sustainability does not have to trade away responsiveness.
  • The framework scales to realistic conditions: 12 global datacenters, 6 heterogeneous node types, 1000 nodes per site, and a workload scaled from the real BurstGPT trace with 0.5x request delays, 3x token counts, and 10x request counts.
  • The workload predictor lets the scheduler react to the rapid request-intensity changes observed in real LLM traces, avoiding stale plans for the next epoch.

Reading between the lines

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

  • The magnitude of the claimed reductions depends entirely on the Section 3 model constants; if real cooling energy is not three times CRAC energy or the blowdown ratio differs, the absolute percentages would shift even if the relative ranking of schedulers does not.
  • A natural extension, not explored in the paper, is feeding renewable-energy and carbon-intensity forecasts into the same metaheuristic to shift load toward cleaner hours, not just cleaner regions.
  • The Pareto-front formulation could be exposed as an operator-facing policy knob, letting a provider set a maximum carbon budget per request and have SLIT pick the best latency within that budget.
  • A real 24-hour testbed deployment with meter readings would be the decisive next check, converting the simulation's 97-99% claims into an operational claim.
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 / 4 minor

Summary. The paper proposes SLIT, a metaheuristic framework that combines a gradient-boosting ML-guided local search with an evolutionary algorithm to schedule LLM inference requests across geo-distributed datacenters. The scheduling objectives are time-to-first-token (TTFT), carbon emissions, water usage, and energy cost. The authors construct synthetic workloads from the BurstGPT trace, model datacenter energy, cooling, water, and carbon behavior in a Python simulator, and compare SLIT-generated plans against Helix and Splitwise. They report very large reductions in carbon, water, energy cost, and TTFT for the single-objective SLIT variants, and a balanced plan that outperforms Helix on all metrics and Splitwise on three of four.

Significance. If the simulation models are trustworthy, the paper addresses a timely and important problem: operational-phase sustainability of LLM inference. The authors should be credited for building on a real LLM request trace, for providing algorithm pseudocode, and for comparing against two published serving systems rather than only toy baselines. The Pareto-front presentation is a useful practical feature. However, all headline numerical claims rest on a simulator whose physical models contain a dimensional error and whose 'validation' is asserted without evidence. The reported 97-99% reductions are therefore not yet evidence of real-world gains.

major comments (4)
  1. [Section 3.3, Eqs. (12)-(13)] The evaporative water model is dimensionally incorrect. Eq. (12) defines W_E,l,t = H_IT,l,t / H_water, where H_water is described as the 'heat capacity of water.' Dividing an energy (J) by a specific heat capacity (J/(kg·K)) yields kg·K, not liters or kilograms of evaporated water. A physically meaningful expression requires the latent heat of vaporization (or the temperature rise across the cooling tower). Since the water savings are a headline result, this error is load-bearing. Relatedly, Eq. (13) defines blowdown as W_B = W_E/(1-D), which diverges as D approaches 1, and no value or source for D is ever given. The water total in Eq. (15) thus depends critically on an unspecified, nonstandard parameter. The water-reduction claims in Section 6 and Fig. 4 should be recomputed with a corrected model and a stated D value, and the sensitivity to D should be reported.
  2. [Section 3.4, Eqs. (16)-(17)] The carbon model contains internal index and notation errors. Eq. (16) writes C_Grid,l,t = CI_l,t × E_tot,t, but E_tot,t has no location index even though the left-hand side is location-specific; it should be E_tot,l,t. Eq. (17) uses W_G,l,t, but the text and Eq. (14) define W_Grid,l,t, not W_G,l,t. These are not merely cosmetic typos because the carbon objective C_tot,t in Eq. (18) is defined as the sum of C_Grid and C_W; the mis-indexing changes the computed carbon footprint whenever datacenters differ in size or grid carbon intensity. The carbon-reduction percentages in Section 6 must be verified after correcting these expressions.
  3. [Section 6, simulator validation and experimental design] The paper states in Section 6, 'We developed and validated a Python-based simulator,' but no comparison to measured energy, water, or carbon data is presented anywhere. The simulator is the sole source of evidence for the claimed reductions. In addition, the experiments are single-run: no error bars, no multiple independent trials, and no sensitivity analysis for key parameters (CoP, D, water intensity, carbon intensity, time-of-use prices, node TDP). Because SLIT's search is stochastic (random initialization, random parent selection, mutation), single-run results cannot support the reported precise percentages. The authors should report multiple seeds with confidence intervals and at least a calibration of the energy and water models against published datacenter measurements or a validated reference model.
  4. [Section 6, Fig. 4 and comparison baselines] Helix and Splitwise are not designed to optimize carbon, water, or energy cost; they optimize serving throughput and latency under heterogeneous-GPU constraints. Comparing a multi-objective sustainability scheduler against them on sustainability metrics therefore makes large reductions almost inevitable and does not by itself demonstrate that SLIT is a better multi-objective optimizer. Also, the paper repeatedly calls the output a 'Pareto optimal' solution set (e.g., Section 1 and Section 6), but a metaheuristic with dominance-based population updates provides no Pareto-optimality guarantee. The authors should either compare against a sustainability-aware baseline or report the baselines' performance on their own objectives, and they should present multi-objective quality indicators (e.g., hypervolume) computed on multiple runs, or restrict the claim to an approximate Pareto set.
minor comments (4)
  1. [Section 3, headings] There are two subsections numbered 3.3 ('Energy Cost Model' and 'Water Model'); the subsequent numbering (3.4 Carbon Model) is consequently misleading.
  2. [References] Reference [27] contains a typo, 'Enviornment' should be 'Environment.'
  3. [Section 6, reproducibility] No table of parameter values is provided for the simulator (e.g., CoP_l, D, WI_l,t, CI_l,t, TOU_l,t, bandwidth, node counts, GPU memory capacities). Without these values, the experiments cannot be reproduced or scrutinized.
  4. [Fig. 3, Algorithm 1] Line 22 of the pseudocode says 'If prediction != observed do' with no specification of how the fallback default plan is generated or how many requests are affected when the predictor is wrong; the impact of prediction error on the reported metrics is not quantified anywhere.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SLIT's reported gains are simulator optimization outcomes, not fitted reductions or self-citation-forced results.

full rationale

The paper's derivation chain is an optimization-plus-simulation study, not a first-principles prediction of physical quantities. The reported reductions (98-99% carbon, 97-99% water, 73-81% TTFT, 96-99% cost) are differences in simulator outputs between SLIT and the Helix and Splitwise baselines. Those outputs are computed from the Section 3 models (Eqs. 5-18); the scheduler minimizes those same objective functions, and the baselines do not, so favorable in-simulator differences are expected but are not circular: no parameter is fitted to the reported outcome, and the Helix/Splitwise comparisons are made under the same models. Model constants are attributed to external references ([22], [23], [24], [25]); Eq. 17's treatment-energy intensities are attributed to prior work [26], a self-citation, but it is a stated model input rather than a load-bearing uniqueness or equivalence argument. The sentence "We developed and validated a Python-based simulator that integrates the models described in Section 3" asserts validation without external measurements, and Eq. 12 has dimensional issues and Eq. 13 leaves D unspecified; these are evidentiary or correctness risks, not circular reductions. No step in the paper defines a predicted quantity as a function of itself, renames a known result, or imports a uniqueness theorem from the authors' prior work. Therefore score 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

SLIT is a software framework, not a postulated physical entity; no new particles, mediators, forces, dimensions, or conserved quantities are introduced. All physical quantities come from cited prior measurements. The free parameters are experimental configuration choices, not physical constants.

free parameters (3)
  • Workload scaling factors = 0.5x delay, 3x token count, 10x request count
    Applied to the BurstGPT trace to create the synthetic workload; chosen by hand to stress the system, with no sensitivity analysis.
  • SLIT hyperparameters = gen, freq, population size X, step size (not specified numerically)
    Control search depth and retraining; no tuning procedure or sensitivity study is reported, so results may depend on these choices.
  • SLIT-Balance weights = normalized equal weights (implied)
    The balanced Pareto solution requires selecting weights to normalize objectives; the exact weighting and selection rule are not specified.
assumptions (5)
  • domain assumption Environmental models in Eqs. 5-18 represent real datacenter energy, cooling, water, and carbon behavior.
    All results are computed from these equations; the simulator's fidelity is asserted, not demonstrated against measurements.
  • domain assumption Cooling energy is 3x CRAC energy and support power is 13% of IT energy (Eqs. 8-9).
    Taken from [23,24]; these fixed factors materially affect energy, carbon, and water totals.
  • domain assumption TTFT is modeled by Eq. 4 as model load plus doubled migration latency plus first-token processing time.
    The latency model ignores queueing, batching dynamics, and GPU sharing at the token level, which are central in LLM serving; this is a strong simplification.
  • domain assumption The next-epoch workload prediction from the linear-regression predictor is accurate enough for scheduling.
    The predictor described in Section 5.1 follows [28]; no prediction error analysis is reported, even though prediction error can trigger the default-plan fallback in the pseudocode.
  • domain assumption Helix and Splitwise are reasonable baselines in this simulator.
    They are repurposed serving systems, not sustainability optimizers; a carbon-aware or water-aware heuristic baseline is absent, so the comparison may overstate SLIT's advantage.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sustainable Carbon-Aware and Water-Efficient LLM Scheduling in Geo-Distributed Cloud Datacenters." pith.science (2026). https://pith.science/paper/RO6XLY7H

@misc{pith2026250523554,
  author       = {Pith},
  title        = {Pith review of: Sustainable Carbon-Aware and Water-Efficient LLM Scheduling in Geo-Distributed Cloud Datacenters},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RO6XLY7H}},
  note         = {Machine review of arXiv:2505.23554}
}
read the original abstract

In recent years, Large Language Models (LLM) such as ChatGPT, CoPilot, and Gemini have been widely adopted in different areas. As the use of LLMs continues to grow, many efforts have focused on reducing the massive training overheads of these models. But it is the environmental impact of handling user requests to LLMs that is increasingly becoming a concern. Recent studies estimate that the costs of operating LLMs in their inference phase can exceed training costs by 25x per year. As LLMs are queried incessantly, the cumulative carbon footprint for the operational phase has been shown to far exceed the footprint during the training phase. Further, estimates indicate that 500 ml of fresh water is expended for every 20-50 requests to LLMs during inference. To address these important sustainability issues with LLMs, we propose a novel framework called SLIT to co-optimize LLM quality of service (time-to-first token), carbon emissions, water usage, and energy costs. The framework utilizes a machine learning (ML) based metaheuristic to enhance the sustainability of LLM hosting across geo-distributed cloud datacenters. Such a framework will become increasingly vital as LLMs proliferate.

Figures

Figures reproduced from arXiv: 2505.23554 by the authors.

Figure 3
Figure 3. Pseudocode for the SLIT framework 5.1 Workload Predictor Due to the high variability of the LLM workload intensity (as observed in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 2
Figure 2. Overview of SLIT framework 5 SLIT Framework Overview Our proposed SLIT framework integrates a metaheuristic approach that utilizes machine learning (ML) assisted local search in combination with an evolutionary algorithm (EA) as shown in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

  1. [1]

    Reducing the Carbon Impact of Generative AI Inference (today and in 2035),

    A. Chien, et al., "Reducing the Carbon Impact of Generative AI Inference (today and in 2035)," in HotCarbon, 2023

    work page 2023

  2. [2]

    OpenAI and the CSU system bring AI to 500,000 students and faculty,

    OpenAI, "OpenAI and the CSU system bring AI to 500,000 students and faculty," 2025. [Online]. [Accessed Apr 2025]

    work page 2025

  3. [3]

    The Unseen AI Disruptions for Power Grids: LLM -Induced Transients,

    Y. Li, et al., "The Unseen AI Disruptions for Power Grids: LLM -Induced Transients," arXiv, 2024

    work page 2024

  4. [4]

    The Environmental Footprint of Data Centers in the United States,

    M.A.B. Siddick, et al., "The Environmental Footprint of Data Centers in the United States," ERL, 2021

    work page 2021

  5. [5]

    Toward a Systematic Survey for Carbon Neutral Data Centers,

    Z. Cao, X. Zhou, H. Hu, Z. Wang, and Y. Wen, "Toward a Systematic Survey for Carbon Neutral Data Centers," ICST, 2022

    work page 2022

  6. [6]

    Future Global Urban Water Scarcity and Potential Solutions,

    C. He, et al., "Future Global Urban Water Scarcity and Potential Solutions," NC, 2021

    work page 2021

  7. [7]

    Data Centre Water Consumption

    D. Mytton, "Data Centre Water Consumption"

    work page

  8. [8]

    Hybrid ant genetic algorithm for efficient task scheduling in cloud data centers,

    M. S. Ajmal, et al., "Hybrid ant genetic algorithm for efficient task scheduling in cloud data centers," CEE, 2021

    work page 2021

Show all 29 references
  1. [9]

    Energy and Network Aware Workload Management for Geographically Distributed Data Centers,

    N. Hogade, et al., "Energy and Network Aware Workload Management for Geographically Distributed Data Centers," TSUSC, 2022

    2022

  2. [10]

    A Survey on Machine Learning for Geo -Distributed Cloud Data Center Management,

    N. Hogade, et al., "A Survey on Machine Learning for Geo -Distributed Cloud Data Center Management," TSUSC, 2023

    2023

  3. [11]

    GreenCourier: Carbon -Aware Scheduling for Serverless Functions,

    M. Chadha, et al., "GreenCourier: Carbon -Aware Scheduling for Serverless Functions," in WoSC, 2023

    2023

  4. [12]

    CASA: A Framework for SLO - and Carbon-Aware Autoscaling and Scheduling in Serverless Cloud Computing,

    S. Qi, et al., "CASA: A Framework for SLO - and Carbon-Aware Autoscaling and Scheduling in Serverless Cloud Computing," in IGSC, 2024

    2024

  5. [13]

    A Framework for SLO, Carbon, and Wastewater -Aware Sustainable FaaS Cloud Platform Management,

    S. Qi, et al., "A Framework for SLO, Carbon, and Wastewater -Aware Sustainable FaaS Cloud Platform Management," in IGSC, 2024

    2024

  6. [14]

    MOSAIC: A Multi -Objective Optimization Framework for Sustainable Datacenter Management,

    S. Qi, et al., "MOSAIC: A Multi -Objective Optimization Framework for Sustainable Datacenter Management," in HiPC, 2023

    2023

  7. [15]

    SWARM Parallelism: Training Large Models Can Be Surprisingly Communication-Efficient,

    M. Ryabinin, et al., "SWARM Parallelism: Training Large Models Can Be Surprisingly Communication-Efficient," in ICML, 2023

    2023

  8. [16]

    Helix: Distributed Serving of Large Language Models via Max - Flow on Heterogenous GPUs,

    Y. Mei, et al., "Helix: Distributed Serving of Large Language Models via Max - Flow on Heterogenous GPUs," arXiv, 2024

    2024

  9. [17]

    Splitwise: Efficient Generative LLM Inference Using Phase Splitting,

    P. Patel, et al., "Splitwise: Efficient Generative LLM Inference Using Phase Splitting," in ISCA, 2024

    2024

  10. [18]

    Hogade, et al., Game-Theoretic Deep Reinforcement Learning to Minimize Carbon Emissions and Energy Costs for AI Inference Workloads in Geo - Distributed Data Centers, TSUSC, 2025

    N. Hogade, et al., Game-Theoretic Deep Reinforcement Learning to Minimize Carbon Emissions and Energy Costs for AI Inference Workloads in Geo - Distributed Data Centers, TSUSC, 2025

    2025

  11. [19]

    BurstGPT: A Real-world Workload Dataset to Optimize LLM Serving Systems,

    Y. Wang, et al., "BurstGPT: A Real-world Workload Dataset to Optimize LLM Serving Systems," arXiv, 2024

    2024

  12. [20]

    Delay -Sensitive Multicast in Inter-Datacenter WAN using Compressive Latency Monitoring,

    T.Y. Cheng, et al., "Delay -Sensitive Multicast in Inter-Datacenter WAN using Compressive Latency Monitoring," ITCC, 2017

    2017

  13. [21]

    Alternating Cold and Hot Aisles Provides More Reliable Cooling for Server Farms,

    R.F. Sullivan, "Alternating Cold and Hot Aisles Provides More Reliable Cooling for Server Farms," WP, 2000

    2000

  14. [22]

    A100 Spec Sheet,

    NVIDIA, "A100 Spec Sheet," 2021. [Online]. Available: https://www.nvidia.com/content/dam/en-zz/Solutions/Data- Center/a100/pdf/nvidia-a100-datasheet-us-nvidia-1758950-r4-web.pdf. [Accessed April 2025]

    2021

  15. [23]

    A survey on data center cooling systems: Technology, power consumption modeling and control strategy optimization,

    Q. Zhang, et al., "A survey on data center cooling systems: Technology, power consumption modeling and control strategy optimization," JSA, 2021

    2021

  16. [24]

    A Review of Data Centers Energy Consumption and Reliability Modeling,

    K. M. U. Ahmed, et al., "A Review of Data Centers Energy Consumption and Reliability Modeling," IEEE Access, 2021

    2021

  17. [25]

    Water Use of Electricity Technologies: A Global Meta -Analysis,

    J. Yi, et al., "Water Use of Electricity Technologies: A Global Meta -Analysis," RSER, 2019

    2019

  18. [26]

    SHIELD: Sustainable Hybrid Evolutionary Learning Framework for Carbon, Wastewater, and Energy -Aware Data Center Management,

    S. Qi, et al., "SHIELD: Sustainable Hybrid Evolutionary Learning Framework for Carbon, Wastewater, and Energy -Aware Data Center Management," in IGSC, 2023

    2023

  19. [27]

    Advanced Weighted Round Robin Procedure for Load Balancing in Cloud Computing Enviornment,

    M. Kushwaha, et al., "Advanced Weighted Round Robin Procedure for Load Balancing in Cloud Computing Enviornment," in Confluence, 2011

    2011

  20. [28]

    Mu: An Efficient, Fair and Responsive Serverless Framework for Resource-constrained Edge Clouds,

    M. Viyom, et al., "Mu: An Efficient, Fair and Responsive Serverless Framework for Resource-constrained Edge Clouds," in ASCC, 2021

    2021

  21. [29]

    Greedy Function Approximation: A Gradient Boosting Machine,

    J.H. Friedman, "Greedy Function Approximation: A Gradient Boosting Machine," TAS, 2001

    2001

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.