PCCN-K (Knowledge Professional) cheat sheet

Cloud computing per NIST has five essential characteristics: on-demand self-service, broad network access, resource pooling, rapid elasticity, and measured service.

Three service models: IaaS supplies compute/storage/network; PaaS supplies runtime and middleware; SaaS supplies a complete application. Control shifts to the provider as you move IaaS to SaaS.

Four deployment models: private, public, hybrid, and community. Hybrid means orchestration between two distinct models, not merely using a public cloud.

Shared responsibility model: provider secures the cloud (physical, hypervisor, facility); customer secures in the cloud (identity, data, configuration) - the split varies by service model.

Service lifecycle stages: strategy, design, transition, operation, and continual improvement. Each stage has distinct outputs and accountable roles.

Utility vs. cloud economics: capital expenditure converts to operating expenditure; pay-per-use billing requires metering, rating, and chargeback capability.

A cloud service catalog is the single source of truth for offered services, tiers, owners, prices, and SLAs; CMDB records deployed configuration items and their relationships.

Multi-tenancy risk: noisy-neighbor contention, data remanence on shared storage, and side-channel exposure. Isolation controls are contractual and technical, not implicit.

Strategy output is a portfolio decision: which services to retire, retain, rehost, replatform, refactor, or repurchase - each with cost, risk, and timeline.

Governance sets decision rights: who approves new cloud services, who owns policy, who accepts residual risk. RACI ambiguity is a governance defect, not an operations problem.

Policy hierarchy: regulation → enterprise policy → cloud control baseline → service configuration. Lower layers must be traceable to a higher requirement.

FinOps operating model: inform (visibility and allocation), optimize (rightsizing, commitment coverage), operate (budgets, anomaly alerts, showback/chargeback).

Vendor lock-in levers: proprietary APIs, data egress cost, identity integration, and licensing. Exit plans must state data export format and maximum extraction time.

Procurement gate: security, legal, privacy, and finance review before a service enters the catalog; exceptions require a named risk owner and expiry date.

Budget variance formula: variance % = (actual − budget) / budget × 100. Positive variance is overspend; report at tag, account, and service level.

Reserved/committed-use coverage = committed spend ÷ total eligible spend. Target coverage is set per workload stability, not uniformly across the estate.

Design for failure: assume any component fails; use redundancy across availability zones, health checks, and automated failover with tested recovery time.

Availability math: serial components multiply (0.99 × 0.99 = 0.9801); redundant parallel paths raise availability as 1 − (1 − A)^n for independent paths.

SLA vs. SLI vs. SLO: SLA is the contractual commitment and remedy; SLI is the measured indicator; SLO is the internal target set tighter than the SLA.

Error budget = 1 − SLO. A 99.9% monthly SLO allows about 43 minutes 50 seconds of unavailability per 30-day month.

Transition includes build, test, deployment, and early-life support. Acceptance criteria must be measurable and signed before handover to operations.

Infrastructure as code gives repeatable, versioned environments; drift detection compares deployed state against declared state and must trigger remediation.

Release vs. deployment: deployment moves code into an environment; release exposes it to users. Decoupling them enables canary and blue/green patterns.

Rollback plan must define trigger thresholds, data-migration reversibility, and maximum tolerable rollback time; irreversible schema changes require a forward-fix plan.

Incident priority derives from impact × urgency, not from who reported it. Severity levels map to response and communication timelines.

Major incident process: declare, assign incident commander, establish communication cadence, restore service first, root-cause analysis after stabilization.

Change types: standard (pre-authorized, low risk), normal (assessed and approved), emergency (expedited with retrospective review).

Problem management removes root causes; incident management restores service. Recurring incidents without a problem record indicate process failure.

Capacity management in cloud: monitor utilization percentiles (p95, p99), not averages; scale on leading indicators such as queue depth or latency.

Backup vs. disaster recovery: backup restores data; DR restores service. RPO defines acceptable data loss; RTO defines acceptable downtime.

Runbooks must be executable by on-call staff without original authors: prerequisites, steps, verification, and escalation contacts.

Access operations: least privilege, just-in-time elevation, periodic entitlement review, and immediate revocation on role change or departure.

Metric types: SLI (measured), KPI (business-relevant performance indicator), and SLA (contractual). Do not report a KPI as if it were an SLA.

Availability % = (agreed total minutes − unplanned downtime minutes) / agreed total minutes × 100. Exclude agreed maintenance windows only if the contract says so.

MTTR = total downtime ÷ number of incidents. MTBF = total uptime ÷ number of failures. Availability ≈ MTBF / (MTBF + MTTR).

Percentile latency: p95 = value below which 95% of requests fall. Averages hide tail latency and are unsuitable for SLA reporting.

SLA credits are remedies, not penalties for the provider's benefit; they are typically tiered by shortfall and often require a customer claim within a window.

Dashboards should show trend plus threshold, with data source and refresh time stated; a metric without an owner and action is decoration.

Cost per transaction = total service cost ÷ transaction volume. Use it to compare tiers and to detect unit-cost drift after scaling.

Measurement integrity: define the measurement point, aggregation window, exclusions, and rounding before agreeing a target; ambiguity becomes a dispute.

Risk = likelihood × impact. Document inherent risk, controls applied, and residual risk; residual risk acceptance requires an accountable owner.

Risk treatment options: avoid, reduce/mitigate, transfer (insurance, contract), and accept. Acceptance must be time-bound and reviewed.

Risk register fields: ID, description, asset/service affected, owner, likelihood, impact, rating, treatment, due date, status.

Continual improvement cycle: measure → analyze gap → plan change → implement → verify effect. Each improvement needs a baseline and a success criterion.

PDCA maps to service management: Plan (strategy and design), Do (transition and operation), Check (measurement), Act (improvement).

Post-incident review is blameless: focus on system conditions and control gaps, produce owned actions with dates, and track them to closure.

Compliance evidence must be current and mapped to the control it satisfies; an expired audit report is a finding, not evidence.

Concentration risk: single provider, single region, or single identity provider dependency. Mitigation is a tested alternative with documented failover time.