Capacity Modeling Applications

Build failure-informed capacity planning models... because "70% utilization" doesn’t mean you’re safe when things break! For years, the go-to logic for network capacity planning has been simple: “If we’re under 70% utilization, we’re good.” But at hyperscale, that’s not just naive: it’s dangerous! Why? Because this model assumes everything works perfectly. It doesn’t account for real-world failure scenarios: fiber cuts, DWDM degradation, hardware failures, or full fault domain isolation. In modern large-scale networks, planning capacity without considering failure is like building a bridge with no load testing. At hyperscale, the question isn't just: “Do we have enough bandwidth?” The real question is: “If we lose two major links, will we still meet SLOs without impacting customer experience?” That’s what failure-informed capacity planning is all about. Some key concepts in failure-informed design: 1. Fault Domains First Before thinking about thresholds, define your fault domains: - Optical paths with shared amplifiers - Racks or rows in the same power/cooling zone - Geographic sites that rely on the same metro transport - Redundant router pairs in the same chassis or fabric Ask yourself: If one fails, what traffic shifts… and where? 2. Traffic Simulations Under Stress We use simulation tools to inject synthetic failure events and answer: - What links/routes absorb the rerouted traffic? - Does anything exceed 100%? - Do any queue depths spike or drop rates increase? - How fast does traffic recover? Simulations don’t predict the future. They pressure-test your assumptions and help you build with confidence. 3. Shadow Traffic Analysis We mirror a subset of real production traffic into test fabrics: - Helps identify unexpected path asymmetries - Surface jitter, delay, or congestion across alternate paths - Validate steering policies before failure happens Think of it as a dress rehearsal for disaster, without affecting live flows. 4. Protective Throttling and Preemption Logic In degraded scenarios, not all traffic is equal. We apply dynamic throttling techniques: - Drop or rate-limit bulk background sync traffic - Preempt non-customer-critical flows - Prioritize payments, voice, and latency-sensitive control-plane sessions Capacity ≠ bandwidth. Real capacity is what remains under fault, not what’s possible during ideal conditions. 5. Automated Headroom Monitors We don’t just track utilization. We monitor available failover capacity: - “Can we absorb the loss of Path A + Path B?” - “What’s our survivable traffic delta under peak load?” - “Has recent growth silently eaten our redundancy?” Dashboards show live survivability margin, not just throughput. What this changed for us: - Avoided multiple potential outages during failover - Validated that certain default ECMP decisions caused localized queue bursts - Tuned BGP and label policies to reroute more gracefully under stress - Helped finance and capacity teams

So you're thinking of building an #electrolyzer to make green #hydrogen. But how much #wind, #solar and #battery capacity do you need to power the electrolyzer in order to minimize the cost of hydrogen it produces? BloombergNEF has just the tool you need to find out - the Hydrogen Electrolyzer Optimization Model (H2EOM). A vastly enhanced version 2.0 was published yesterday by my brilliant colleagues Xiaoting Wang and Ulimmeh-Hannibal Ezekiel. For an example project in #California, the optimal setup for a 1MW electrolyzer is to power it by 1.14MW of wind and 0.83MW of solar, skipping the batteries. That gives you a levelized cost of hydrogen (LCOH) or $4.63 per kilogram and a utilization rate of 65% on your electrolyzer (excluding any #IRA #45V #taxcredits). If you wanted to increase the utilization rate to 90%, you'd need to be happy with a #LCOH of $7.28 per kilogram as you pay for batteries, as well as more solar and wind capacity. Users can do this modeling for any location on the planet by using BNEF's Solar- and Wind Capacity Factor Tool to get 8,760h of capacity factor data anywhere. Users can tweak any cost and financing assumption to suit their project, making this a super versatile tool for #H2 modeling. Oh, and did I say you can model up to 50 projects at once? BNEF clients can download the model here: https://lnkd.in/e9vTYc7G

Publicly Accessible Energy Storage Systems (ESS) Simulation Price-taker models are suitable for small-scale ESS as their capacity does not influence market prices or system dispatch. This post highlights DOE price-taker valuation tools. 🟦 1) QuESt  QuESt is a free, open-source Python application suite for energy storage simulation and analysis, developed at Sandia National Laboratories. It includes three interconnected applications:  1- QuESt Data Manager,  2-QuESt Valuation, and  3-QuESt BTM, Eligible technologies include BESS (Li-ion, advanced lead-acid, vanadium redox), flywheels, and PV, using a shared model for different BESS and flywheel types based on their parameters. 🟦 2) Renewable Energy Integration and Optimization (REoptTM)  The REopt™ platform, developed by the National Renewable Energy Laboratory (NREL), optimizes energy systems for various applications, recommending the best mix of renewable energy, conventional generation, and energy storage to achieve cost savings, resilience, and performance goals. Eligible technologies include: PV, wind, CHP, electric and thermal energy storage, absorption chillers, and existing heating and cooling systems. 🟦 3) Distributed Energy Resources Customer Adoption Model (DER-CAM)  DER-CAM is a decision support tool from Lawrence Berkeley National Laboratory (LBNL) designed to optimize DER investments for buildings and multienergy microgrids. Eligible technologies include conventional generators, CHP units, wind and solar PV, solar thermal, batteries, electric vehicles, thermal storage, heat pumps, and central heating and cooling systems. 🟦 4) System Advisor Model (SAM) SAM is a techno-economic computer model that evaluates the performance and financial viability of renewable energy projects. It includes performance models for various systems such as PV (with optional battery storage), concentrating solar power, solar water heating, wind, geothermal, and biomass, and a generic model for comparison with conventional systems. Eligible technology types focus on electrochemical ESS, supporting lead-acid, Li-ion, vanadium redox flow, and all iron flow batteries. Users can also model custom battery types by specifying their voltage, current, and capacity. SAM offers detailed modelling of battery cells, power converters, and factors like degradation, voltage variation, and thermal properties. 🟦 5) Energy Storage Evaluation Tool (ESETTM) ESETTM is a suite of modules developed at PNNL that allows utilities, regulators, and researchers to model and evaluate various ESSs. ESETTM features a modular design for ease of use and currently includes five modules for different ESS types, such as BESSs, pumped-storage hydropower, hydrogen energy storage, storage-enabled microgrids, and virtual batteries. Some applications also include distributed generators and photovoltaics (PV). Source: see post image. Link to the modellers: in the comment section This post is for educational purposes only.

The Fab Whisperer: Capacity Planning - From Spreadsheets to Self-Learning Models. Last week we looked at the widening gap between silicon demand and fab capacity — the classic setup for another boom-and-bust cycle. Imbalance is inherent in the market. We try to balance it in the way we plan capacity. For an industry that spends hundreds of billions on CAPEX, capacity planning should be science. Yet it often I see frozen spreadsheets, heroic assumptions, and “best-guess” throughput models that quietly drift from reality. Are we building fabs based on models that no longer represent how fabs actually run? Using the wrong model for the wrong purpose? CAPEX Planning ≠ Fab Daily Operations Planning Capacity — deciding what, when, and where to build. Running Capacity — managing flow, bottlenecks, and daily WIP. CAPEX models are strategic: they test economics, demand scenarios, and sensitivity to capacity detractors. Operational models are tactical: they simulate variability, queueing, and dispatch logic. When fabs try to use the same model for both, they end up with bad investments and bad daily decisions. It’s like using a telescope to check your pulse. Most Common Methods of How We Plan Capacity 1. Static Models (Spreadsheet Economics) Quick and transparent — perfect for early CAPEX justifications. But fixed throughput and yield assumptions age fast. Once products, recipes, or WPH shift, the model collapses. 2. Dynamic Simulations (Discrete-Event or Digital Twins based) Capture queues, PM downtime, and rework loops — essential for operational decision-making. Great for optimizing how to run a fab, not what to build next. Powerful but maintenance-heavy; too often abandoned after the big study. The Next Frontier Not mainstream yet but they point to the future: AI-Driven and Hybrid Models. These models will learn from real time fab data, adapt to product mix, and continuously recalibrate effective capacity. They will bridge the gap between planning and operations — a single living model that never goes stale. The barrier isn’t technology — it’s data discipline and trust. The Real Challenge The biggest risk isn’t model complexity — it’s model decay. Assumptions age. Routings evolve. PM cycles shift. By the time the next CAPEX round starts, you’re planning the future based on a fab that no longer exists. What can we do meanwhile Match the model type to the decision horizon. CAPEX → financial sensitivity and long-term. Operations → flow dynamics, variability control, short term. Treat models as living systems, not one-off projects. Assign ownership for keeping assumptions, routings, and rates current. Benchmark quarterly — compare modeled vs. actual effective capacity. Start building the bridge: integrate AI and fab data into planning cycles today. Are your capacity models describing reality — or nostalgia? #TheFabWhisperer #Semiconductor #FabOperations #CapacityPlanning #DigitalTwin #AI #ManufacturingExcellence #FabModeling

Capacity Plan Modeling pt 2 - BPO vs. Captive My last post was about modeling in general, and I wanted to go into some details that can be used for different situations: BPO vs. Captive Centers. This is an interesting topic because while the overall approach is generally similar, the goals can be very different while the details of what you're doing is mostly the same. Captive Centers: the approach here will depend on the model goal. Goals typically will be (but not exclusively these) cost reduction, hours of operation, SLA changes, new business or line of business, occupancy changes, and site alignment. If the Captive Center has anything outsourced or is considering it, most of these models will want to keep that in mind for alignment. New LOB, Cost & Quality are among the top reasons for running these models, so being able to clearly communicate the changes & why they're recommended are very important. Per the last post, we want to also have clear goals & expected end results. BPOs: the approach here is usually new business or expansion of existing business. Goals typically will be (but not exclusively these) ramp for peak season, ramp down for post peak or loss of business, pricing changes, hours of operation/site alignment changes, and new business or new line of business. As a result, a solid understanding of the business as well as how the model can be as efficient as possible is vital since margins are so tight. When creating & evaluating the models, know the labor laws (restrictions & opportunities), site/region cost structure, cost/agent & the load ratio, concurrent connections for remote staff, ratios for supervisor/QA/Trainer/etc., the potential for promotions of temp or permanent staff to these levels, and the available recruiting & TA resources for cross-training or ramp needs. Finally, show more than one result. Leadership appreciates being able to have options, and if your options show benefits, opportunities, risks & mitigation strategies for each, you'll establish your credibility in the business and become a trusted advisor. Also consider the need to show options that enable the business to adjust quickly based on unexpected business changes that may change the environment quickly.

𝗦𝗺𝗮𝗿𝘁 𝗖𝗮𝗽𝗮𝗰𝗶𝘁𝘆 𝗔𝗹𝗹𝗼𝗰𝗮𝘁𝗶𝗼𝗻 𝗦𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗲𝘀 𝗳𝗼𝗿 𝗗𝗲𝗰𝗲𝗻𝘁𝗿𝗮𝗹𝗶𝘇𝗲𝗱 𝗔𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀 𝗶𝗻 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁 𝗙𝗮𝗯𝗿𝗶𝗰 (Part 2/4) 𝗦𝗰𝗮𝗹𝗶𝗻𝗴 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁 𝗙𝗮𝗯𝗿𝗶𝗰 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 in a decentralized self-service analytics environment requires balancing two critical goals: 1. 𝗖𝗼𝗻𝘀𝗼𝗹𝗶𝗱𝗮𝘁𝗶𝗼𝗻 & 2. 𝗜𝘀𝗼𝗹𝗮𝘁𝗶𝗼𝗻.  With many teams using Fabric simultaneously, capacity 𝗮𝗱𝗺𝗶𝗻𝘀 𝗺𝘂𝘀𝘁 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝘇𝗲 𝘁𝗼 𝗺𝗮𝘅𝗶𝗺𝗶𝘇𝗲 𝗿𝗲𝘀𝗼𝘂𝗿𝗰𝗲 𝘂𝘁𝗶𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 without compromising performance. 𝗖𝗮𝗽𝗮𝗰𝗶𝘁𝘆 𝗮𝗹𝗹𝗼𝗰𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀 𝗶𝗻 𝗮 𝗺𝘂𝗹𝘁𝗶𝘁𝗲𝗮𝗺 𝗲𝗻𝘃𝗶𝗿𝗼𝗻𝗺𝗲𝗻𝘁 𝗗𝗲𝗱𝗶𝗰𝗮𝘁𝗲𝗱 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 per department/domain • 𝗘𝗮𝗰𝗵 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝘂𝗻𝗶𝘁, like Finance or Marketing, gets its own Fabric capacity, often managed through separate Azure subscriptions or resource groups. • 𝙋𝙧𝙤𝙨: Complete isolation, clear cost attribution, guaranteed performance for mission-critical workloads. • 𝘾𝙤𝙣𝙨: Risk of idle, underused capacity 𝗹𝗲𝗮𝗱𝗶𝗻𝗴 𝘁𝗼 𝗵𝗶𝗴𝗵𝗲𝗿 𝗰𝗼𝘀𝘁𝘀; increased administrative overhead managing multiple capacities. 𝗦𝗵𝗮𝗿𝗲𝗱 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 across departments (consolidation) • Multiple teams share a 𝗹𝗮𝗿𝗴𝗲𝗿 𝗽𝗼𝗼𝗹 𝗼𝗳 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆, which smooths peak demands and reduces idle resources. • 𝙋𝙧𝙤𝙨: Higher overall utilization, cost efficiency, enables cross-team collaboration, and often reduces licensing costs. • 𝘾𝙤𝙣𝙨: Risk of one team’s heavy usage impacting others (“noisy neighbor” problem), requiring strong governance and monitoring. 𝗛𝘆𝗯𝗿𝗶𝗱 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵: Best of both worlds Organizations often combine both models. For example, light workloads might share capacity, while mission-critical or heavy workloads get dedicated capacity. This flexible approach adapts as teams grow and usage patterns change. 𝗣𝗹𝗮𝗻𝗻𝗶𝗻𝗴 𝗰𝗼𝗻𝘀𝗼𝗹𝗶𝗱𝗮𝘁𝗶𝗼𝗻 𝗮𝗻𝗱 𝗰𝗵𝗮𝗿𝗴𝗲𝗯𝗮𝗰𝗸 • 𝗨𝘀𝗲 𝘀𝗵𝗮𝗿𝗲𝗱 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 𝗳𝗼𝗿 𝗺𝘂𝗹𝘁𝗶𝗽𝗹𝗲 𝘀𝗺𝗮𝗹𝗹 𝗼𝗿 𝗺𝗶𝗱-𝘀𝗶𝘇𝗲𝗱, noncritical workloads to maximize ROI. • 𝗔𝘀𝘀𝗶𝗴𝗻 𝗱𝗲𝗱𝗶𝗰𝗮𝘁𝗲𝗱 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 𝗳𝗼𝗿 𝗹𝗮𝗿𝗴𝗲𝗿 𝗼𝗿 𝗰𝗿𝗶𝘁𝗶𝗰𝗮𝗹 𝘄𝗼𝗿𝗸𝗹𝗼𝗮𝗱𝘀 needing guaranteed performance. • 𝗟𝗲𝘃𝗲𝗿𝗮𝗴𝗲 𝘁𝗵𝗲 𝗙𝗮𝗯𝗿𝗶𝗰 𝗖𝗵𝗮𝗿𝗴𝗲𝗯𝗮𝗰𝗸 𝗮𝗽𝗽 to transparently attribute capacity usage by team, promoting accountability and responsible consumption. • 𝗖𝗼𝗻𝘀𝗶𝗱𝗲𝗿 𝗳𝗮𝗰𝘁𝗼𝗿𝘀 𝗹𝗶𝗸𝗲 𝗴𝗲𝗼𝗴𝗿𝗮𝗽𝗵𝗶𝗰 𝗹𝗼𝗰𝗮𝘁𝗶𝗼𝗻, data domain, workload patterns, and service-level agreements when grouping workloads. References : https://lnkd.in/geYuFJq8 #MicrosoftFabric #DataEngineering #DataAnalytics #PowerBI  #DataPlatform #FabricCommunity #MicrosoftLearn #CapacityPlanning