OpenDC-STEAM: Realistic Modeling and Systematic Exploration of Composable Techniques for Sustainable Datacenters Technical Report
Dante Niewenhuis, Sacheendra Talluri, Alexandru Iosup, Tiziano De Matteis · 2026
Read OpenDC-STEAM: Realistic Modeling and Systematic Exploration of Composable Techniques for Sustainable Datacenters Technical Report by Dante Niewenhuis, Sacheendra Talluri, Alexandru Iosup, Tiziano De Matteis (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-12381.txt.
Introduction
Niewenhuis, Talluri, Iosup, and de Matteis build OpenDC-STEAM, a simulator that tests green datacenter techniques. Such tricks are often judged alone in small setups that miss real dynamics. Their open tool checks single and combined effects on carbon output and speed. Three techniques face varied workloads and site setups: horizontal scaling, batteries, and temporal shifting. Mixing strategies can cut emissions sharply, at the cost of trade-offs the tool helps users weigh.
Catalog record
| CREATOR | Dante Niewenhuis, Sacheendra Talluri, Alexandru Iosup, Tiziano De Matteis |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62513 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.12381
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
OpenDC-STEAM: Realistic Modeling and Systematic Exploration of Composable Techniques for Sustainable Datacenters Technical Report Dante Niewenhuis, Sacheendra Talluri, Alexandru Iosup, Tiziano De Matteis Abstract The need to reduce datacenter carbon-footprint is urgent. While many sustainability techniques have been proposed, they are often evaluated in isolation, using limited setups or analytical models that overlook real-world dynamics and interactions between methods. This makes it challenging for researchers and operators to understand the effectiveness and trade-offs of combining such techniques. We design OpenDC-STEAM, an open-source customizable datacenter simulator, to investigate the individual and combined impact of sustainability techniques on datacenter operational and embodied carbon emissions, and their trade-off with performance. Using STEAM, we systematically explore three representative techniques–horizontal scaling, leveraging batteries, and temporal shifting–with diverse representative workloads, datacenter configurations, and carbon-intensity traces. Our analysis highlights that datacenter dynamics can influence their effectiveness and that combining strategies can significantly lower emissions, but introduces complex cost-emissions-performance trade-offs that STEAM can help navigate. STEAM supports the integration of new models and techniques, making it a foundation framework for holistic, quantitative, and reproducible research in sustainable computing. Following open-science principles, STEAM is available as FOSS: https://github.com/atlarge-research/OpenDC-STEAM . This is an extended version of a paper published at CCGRID 2026. I Introduction The Information and Communication Technology (ICT) sector accounts for an estimated 1.5%-4% of global carbon emissions [ 4 , 14 ] , with datacenters contributing for a large and growing share [ 1 , 35 ] . Policy mandates, user expectations, and financial considerations are pushing the computing community to control and reduce datacenter carbon footprint [ 18 , 11 ] . In response, many techniques have been proposed to reduce the carbon emissions of a datacenter, including load shifting [ 2 , 30 ] , hardware scaling [ 20 , 33 ] , or batteries [ 2 , 28 , 19 ] . However, these strategies are often evaluated using analytical models or limited setups that do not account for interactions between sustainability techniques and with real-world datacenter dynamics, such as task failures. Their effectiveness also depends on multiple factors, such as the scale of the datacenter, workload characteristics, the energy source carbon intensity , and the embodied carbon emitted during hardware manufacturing [ 30 , 24 ] . In such a broad design space, non-trivial trade-offs in costs, carbon emissions, and performance frequently arise. To navigate this complexity and make informed decisions on how to design datacenters and configure their operations, datacenter designers, operators, and sustainability researchers need quantitative insights . In this work, we introduce OpenDC-STEAM (STEAM in the following), an open-source simulation-based tool for quantifying the effects of sustainability techniques on datacenter performance and carbon footprint. Fig. 1 : STEAM quantifies the impact and trade-offs of sustainability techniques. In the figure, (A) results for the Surf workload are shown as total carbon reduction [%], (B) peak power draw [kW], and (C) average task delay [h]. Stars indicate the best performing technique for each metric. HS: Horizontal Scaling, TS: Temporal Shifting, B: Batteries. STEAM is a new capability-driven feature in the OpenDC space [ 23 ] . STEAM’s composable architecture allows users to analyze trade-offs arising from combining techniques, evaluate new techniques, and tune existing ones. Its flexibility and performance enable datacenter designers, operators, and researchers to rapidly explore a wide range of ”what-if” scenarios and gain actionable insights into sustainability decisions. We use STEAM to quantitatively evaluate three representative sustainability techniques–horizontal scaling, temporal shifting, and batteries–on datacenter performance (e.g., service quality) and carbon footprint, considering diverse and realistic workloads, carbon regions, datacenters, and their dynamics. We show that each technique can reduce carbon emissions with minimal performance impact, but the actual extent of this reduction depends on the workload, available hardware, location, and operational phenomena such as task failures. We highlight how disregarding datacenter dynamics has led to optimistic carbon emissions reductions in prior work. Figure 1 illustrates how STEAM can be used to understand the impact of sustainability techniques. Combining techniques can further improve their effectiveness (see Figure 1 A), but also introduces trade-offs, such as power-draw spikes ( Figure 1 B) and service-quality degradation ( Figure 1 C), which STEAM can quantify. With STEAM, we contribute to the efforts towards better understanding modern computing ecosystems [ 17 ] . STEAM provides a framework for the systematic evaluation of sustainability techniques, helping datacenter providers and researchers to make evidence-based decisions on datacenter design and operation. This work makes the following contributions: 1. We motivate and design STEAM 1 1 1 https://github.com/atlarge-research/OpenDC-STEAM ( \lx@sectionsign III and \lx@sectionsign IV ). STEAM is an open-source, flexible framework for quantifying the effect of sustainability techniques on the datacenter carbon footprint and performance. 2. Using STEAM through a relevant experimental setup ( \lx@sectionsign V ), we conduct an extensive analysis of three representative sustainability techniques, showing that: (a) Horizontal scaling can reduce carbon footprint (up to 35%), but ignoring operational phenomena, such as failure, leads to optimistic results ( \lx@sectionsign VI-A ). (b) Batteries can significantly improve the sustainability of datacenters (up to 28%) when used in the right context, but can also reduce it when misused ( \lx@sectionsign VI-B ). (c) Temporal shifting is less effective (7%) than shown by previous work when operational effects, such as stacking tasks, are taken into account ( \lx@sectionsign VI-C ). 3. We leveraged STEAM’s composable design to conduct a first-of-its-kind analysis of using multiple sustainability techniques in conjunction ( \lx@sectionsign VII ). Combining techniques can increase effectiveness but add new, complex trade-offs that STEAM helps quantify and explain. II Background This section introduces the concepts and techniques that enable realistic modeling of sustainability techniques in STEAM. II-A Workload A workload contains all jobs executed by a datacenter in a specific period. We use a bag-of-tasks workload model where users submit jobs to a datacenter, each containing one or more tasks. Workloads are defined using traces containing the arrival times of all tasks, resource requirements, durations, deadlines, and sampled resource utilization over time. Workloads are often created by collecting operation logs of existing datacenters. II-B Carbon Emissions Carbon emissions are all the CO2 and CO2-equivalent of other greenhouse gases emitted by a datacenter. We distinguish between operational and embodied carbon emissions. Operational Carbon refers to all emissions generated from producing energy used to power servers, cooling systems, and other infrastructure in a datacenter. The type of energy a datacenter uses significantly impacts its operational carbon footprint, with coal-based electricity emitting up to 20x more carbon than renewable sources [ 16 ] . Since green energy availability fluctuates [ 30 ] , accurate estimation requires information about the carbon intensity over time, often provided as traces from services such as ElectricityMaps 2 2 2 https://www.electricitymaps.com/ or Watttime 3 3 3 https://watttime.org/ . Embodied Carbon considers emissions from manufacturing datacenter components such as CPUs and GPUs. As the energy efficiency of datacenters improves, embodied carbon has become a major contributor to the overall carbon footprint [ 22 ] . A common method to calculate the embodied carbon of a component during a workload is to multiply the total embodied carbon cost of the component by the portion of the component’s lifetime the workload used [ 16 ] . II-C Sustainability Techniques