PCAM key terms

A structured set of design principles and best-practice pillars (operational excellence, security, reliability, performance efficiency, cost optimization, sustainability) used to evaluate workloads and identify improvement opportunities before and after deployment.

A short document capturing one significant design choice, the context that forced it, the alternatives considered, and the consequences accepted, so future teams understand why the system looks the way it does.

A pre-provisioned, governed multi-account or multi-subscription environment with baseline identity, networking, logging, and guardrails, giving new workloads a compliant starting point instead of building controls from scratch each time.

An automated pipeline that creates new cloud accounts or subscriptions from a template, applying baseline configuration, network attachment, identity roles, and logging so provisioning is repeatable and auditable.

An automated check that continuously verifies an architectural characteristic such as latency budget, dependency direction, or encryption coverage, turning design intent into a test that fails the pipeline when violated.

A logically isolated network boundary in which you define IP ranges, subnets, route tables, and gateways, controlling which resources can communicate with each other and with external networks.

A regional network hub that connects many VPCs, on-premises links, and shared services through a single attachment point, replacing a mesh of pairwise peering connections with centralized routing and inspection.

A network design where a central hub hosts shared connectivity, inspection, and egress, and spoke networks connect only through it, simplifying policy enforcement and reducing peering complexity.

A network interface placed inside your subnet that maps to a managed service, letting traffic reach that service over private IP addresses and avoiding exposure to the public internet.

A DNS namespace resolvable only from associated networks, used to map a managed service's public hostname to its private endpoint address so existing client code works unchanged.

Control over outbound traffic leaving a network, typically enforced by a central inspection VPC, firewall appliances, or proxy, so compromised workloads cannot exfiltrate data or reach untrusted destinations.

Dividing a network into zones with distinct trust levels and explicit allow rules between them, limiting lateral movement if one zone is compromised and making traffic flows easier to audit.

Granting an identity only the permissions needed for its actual task, for only as long as needed, so a leaked credential or misused role cannot affect unrelated resources.

Obtaining temporary credentials for a role rather than using long-lived keys, typically after a trust policy check, which enables short sessions, audit trails, and cross-account access without shared secrets.

Trusting an external identity provider to authenticate users and assert their attributes, so the cloud platform issues sessions based on those assertions instead of maintaining separate user passwords.

A maximum-permission envelope attached to an identity that caps what its policies can grant, letting teams self-manage roles without being able to exceed the boundary set by a central team.

An organization-level guardrail that sets the maximum available permissions for accounts or organizational units, blocking actions such as disabling logging regardless of what local administrators allow.

Authorization that evaluates tags or attributes of the principal, resource, and request context, allowing rules like matching cost center or environment instead of enumerating individual resource names.

Expressing governance rules in a machine-readable form that is version-controlled, tested, and enforced automatically in pipelines and at runtime, replacing manual review with repeatable checks.

Moving objects between storage classes with different retrieval latency and cost as their access frequency changes, often driven by lifecycle rules, to reduce spend on cold data.

A rule set that automatically transitions, expires, or deletes stored data after defined ages or access patterns, enforcing retention requirements and preventing unbounded storage growth.

A centralized repository holding raw and refined data in open formats across many sources, queried by multiple engines, typically organized into landing, curated, and consumption zones.

Physically splitting a dataset by a low-cardinality key such as date or region so queries scan only relevant subsets, improving performance and reducing cost per query.

Defined rules for how long each class of data is kept and when it must be destroyed, driven by legal, contractual, and operational needs and enforced by automated lifecycle mechanisms.

Incrementally replacing a legacy system by routing a growing share of functionality to new services while the old system still runs, allowing rollback and avoiding a single risky cutover.

An isolation layer that translates between a legacy system's model and a new service's model, preventing legacy concepts and coupling from leaking into the modernized codebase.

Moving an application to managed cloud services with limited code change, such as replacing a self-managed database with a managed one, capturing operational benefits without a full rewrite.

A mechanism by which a client locates a healthy instance of a service by name, using a registry or DNS, so instances can scale, move, or fail without hard-coded addresses.

The maximum acceptable amount of data loss measured in time, which determines how frequently backups or replication must occur for a workload.

The maximum acceptable elapsed time to restore a workload after disruption, which drives decisions about standby capacity, automation level, and failover runbooks.

Continuously copying data and artifacts to a second region so a workload can resume there after a regional failure, with tradeoffs in cost, latency, and consistency.

A resilience pattern that stops sending requests to a failing dependency after repeated errors, failing fast and allowing the dependency time to recover instead of amplifying load.

A tested, step-by-step procedure for shifting traffic and data to a standby environment, including decision criteria, validation checks, and rollback steps, so recovery does not depend on improvisation.

Deliberately injecting failures such as instance loss or network delay in controlled experiments to verify that resilience mechanisms work and assumptions about failure modes hold.

The full cost of running a workload over time, including compute, storage, network transfer, licensing, support, and staff effort, used to compare architectures fairly rather than by list price alone.

Matching provisioned capacity to observed demand by analyzing utilization metrics and adjusting instance sizes or counts, eliminating both over-provisioning waste and under-provisioning risk.

A reduced rate obtained by committing to a defined amount of usage over a term, appropriate for steady predictable workloads, in exchange for lower flexibility than on-demand pricing.

Automatically adding or removing capacity in response to demand signals, keeping performance within target while reducing cost during idle periods.

Defining infrastructure in version-controlled declarative files that a tool applies idempotently, enabling review, repeatable environments, and drift detection instead of manual console changes.

Comparing the live configuration of deployed resources against the declared desired state and reporting differences, so unauthorized or accidental manual changes are surfaced and corrected.

A property of data or infrastructure that prevents modification after creation, such as write-once storage or replaced-not-patched servers, supporting both recovery and audit integrity.

A runtime switch that enables or disables functionality without redeploying, allowing progressive rollout, instant rollback, and separation of release from deployment.

Routing a small percentage of traffic to a new version while monitoring error and latency signals, then gradually increasing exposure or rolling back automatically if thresholds are breached.