C2PF · from objectives to architecture4 layers01InputDeclared objectives · SLO · PLO · RLO02EvaluationClassification vs. thresholds (SLA + SRE)03DecisionSmallest class per objective · keep the largest04OutputArchitectural plan + rationale→ objectives turned into traceable architecture
Research · Cloud · Capacity Planning

C2PF (Cloud Capacity Planning Framework)

A scientific core for planning cloud capacity up front — from declared objectives to architecture.

Carlos Diego Cavalcanti Pereira · Doctoral Thesis, CESAR School (2023)

  • 4 decision layers
  • SLO · PLO · RLO
  • CESAR thesis · 2023
  • 2 papers + product

Architectural decisions in cloud capacity planning rarely derive from explicit, structured objectives. C2PF closes that gap.

From thesis to executable instance
Inside the framework

Meta-model, classes and a capacity calculator

The framework subpage details the nine meta-model elements, the 50 specifications of each c2pf1 class and the general equation, and includes a calculator that turns a service's objectives into a class, a plan by layer and a downloadable rationale.

The problem

Sizing the cloud by precedent — with no measure of accuracy

The industrial paradigm plans capacity from historical usage precedents — and never establishes how accurate those estimates are. For workloads without history, the method simply does not apply.

At the architectural level, decisions — which services to provision, at what reliability, under what performance envelope — rarely derive from explicit, structured objectives. Existing methods either presuppose a prior architecture (ATAM, CBAM) or produce generic outputs disconnected from cloud-specific constructs (ADD, well-architected frameworks). And at the assessment stage, everything depends on tacit knowledge: two architects facing the same problem arrive at different proposals — and that does not scale.

Motivation

The risk lives at the start of the cycle

Cloud projects follow a familiar cycle: assessment, sizing, migration and operation/optimization. It is at the start — assessing and sizing — that the risk and cost of decisions concentrate, and that is exactly where the least rigor exists. Planning that only looks backward cannot justify its own accuracy nor handle the unprecedented.

The proposal is not another framework or another AI tool, but a scientifically grounded decision pipeline: turning declared objectives into architecture traceably, reproducibly and cloud-specifically — and, from there, externalizing specialist knowledge into platform-executable structures.

Research & solution

From objective to architecture, in four layers

C2PF is a formal classification model for cloud workloads via Service-Level (SLO), Performance-Level (PLO) and Reliability-Level (RLO) objectives, with abstract sizing classes. A decision-oriented process model closes the instantiation gap — how to derive an architectural decision from the declared values:

01 · Input

Structured elicitation

Collects SLO, PLO and RLO declarations in a structured format.

02 · Evaluation

Threshold classification

Evaluates each dimension against discrete bands grounded in SLA tiers and SRE practice.

03 · Decision

Smallest class per objective + overrides

Computes the smallest class that meets each objective type and takes the largest of the three. Under a budget cap it relaxes the PLO, then the SLO, never the RLO; compliance and compute enter as overrides.

04 · Output

Traceable plan

Generates the architectural plan with explicit decision rationale.

  • ›Five principles: reproducibility, traceability, hierarchy-driven conflict resolution, domain specificity and conservative boundary resolution.
  • ›Assessment Model (5 phases): externalizes assessment knowledge into reproducible, executable structures — each elicitation question maps to a C2PF capacity characteristic.
  • ›Constrained LLM synthesis: the language model only translates a machine-readable assessment profile into a Statement of Work (SOW) — it does not invent sizing.
Intended results

Capacity decisions as an engineering artifact

Make design-time capacity decisions traceable to declared objectives, reproducible and cloud-specific; externalize tacit assessment knowledge so it scales across practitioners; and open the path from intent to executable, validated architecture.

Advances already demonstrated

From thesis to commercial operation

C2PF began in the doctoral thesis (CESAR School, 2023) and evolved into two papers and a product in production:

1–2 daysproposal cycle, against 5–15 business days in the internal baseline, over three months of commercial operation.
81%of contacted clients (42 of 52) advanced to a second meeting.
2 of 82IaC commits changed sizing in the thesis industrial application.
  • ›End-to-end example with a B2C payment service under PCI DSS, where the declared throughput exceeds the L envelope and leads to class c2pf2.xlarge; compliance controls are added without changing the class.
  • ›A challenge repository covering 6 industries, 30 categories and ~350 questions of assessment with pre-defined sizing characteristics.
  • ›The product in use shows that scientific abstraction, properly engineered, dramatically reduces time-to-proposal.
Next steps

Where the research goes next

  • 01Real telemetry for the simulator. The five scenarios and the 27-cell sweep were evaluated in simulation; calibrating with production data is the next step.
  • 02From intent to executable, validated architecture — turning the SOW (a transition artifact) into a formal, executable architectural decision.
  • 03Blind comparison between model-generated SOWs and architect-written SOWs, without the mandatory review in between.
Research & development opportunities

Open fronts for collaboration

  • ›Executable architecture — couple the C2PF plan to Infrastructure-as-Code (IaC) generation.
  • ›Cost/FinOps optimization — link the sizing classes to cost models.
  • ›Repository expansion — more verticals, categories and capacity characteristics.
  • ›Validation instruments — benchmarks and protocols for design-time capacity planning.
  • ›Domain extension — data and AI, beyond cloud, in the same capacity grammar.
  • ›Sector applications — migration, data platforms and AI adoption in professional services.
Publications & research

Read the science behind C2PF

Doctoral thesis (CESAR School, 2023), a chapter (IntechOpen, 2025) and the papers that instantiate the framework — from the process model to LLM-driven assessment.

Cloud Capacity Planning Framework · C2PF · CESAR School · MIT Sloan