Cloud Computing Cloud Architecture and Design Syllabus

What is an Availability Zone (AZ)?

One or more discrete, isolated data centers within a region, each with independent power, cooling, and networking, connected via low-latency links. Multi-AZ deployment protects against data-center-level failures.

What is the difference between Multi-AZ and Multi-Region deployment?

Multi-AZ spreads resources across isolated data centers within one region for high availability and low-latency failover. Multi-Region spans geographically separate regions for disaster recovery, lower global latency, and protection against region-wide outages, but adds complexity and data-replication cost/latency.

Why does cross-region replication introduce challenges that multi-AZ does not?

Regions are far apart, so replication incurs higher network latency, data-consistency tradeoffs (often eventual consistency), higher cost, and data-sovereignty/compliance considerations.

What are the four common Disaster Recovery (DR) strategies, ordered by cost and recovery speed?

1) Backup and Restore (cheapest, slowest), 2) Pilot Light, 3) Warm Standby, 4) Multi-Site Active/Active (most expensive, fastest). Speed and cost increase from backup/restore to active/active.

Define RTO (Recovery Time Objective).

The maximum acceptable time to restore a system to operation after a disruption — i.e., how long you can afford to be down.

Define RPO (Recovery Point Objective).

The maximum acceptable amount of data loss measured in time — i.e., how far back in time recovery data must reach (how much recent data you can afford to lose).

Contrast the Pilot Light and Warm Standby DR strategies.

Pilot Light keeps only core services (e.g., replicated database) running minimally in the DR site, scaling up the rest on failover. Warm Standby runs a scaled-down but fully functional copy always on, giving faster recovery at higher cost.

In DR strategies, which has the lowest RTO/RPO: backup-and-restore or multi-site active/active?

Multi-site active/active has the lowest (near-zero) RTO and RPO because a fully running duplicate serves traffic continuously; backup-and-restore has the highest RTO/RPO.

What is a monolithic architecture?

An application built as a single, unified, tightly-coupled deployable unit where all functionality (UI, business logic, data access) runs in one process/codebase and is deployed together.

What is a microservices architecture?

An architectural style structuring an application as a suite of small, independently deployable services, each owning a single business capability and communicating over lightweight protocols (e.g., HTTP/REST, messaging).

What is the Strangler Fig pattern for monolith-to-microservices migration?

Incrementally replacing pieces of a monolith by routing specific functionality to new microservices behind a facade/proxy, gradually 'strangling' the monolith until it can be retired — avoiding a risky big-bang rewrite.

Name two advantages and one disadvantage of microservices over monoliths.

Advantages: independent deployment/scaling of services and technology heterogeneity (fault isolation). Disadvantage: increased operational complexity — distributed systems bring network latency, data consistency challenges, and harder debugging/monitoring.

What is service decomposition, and by what principle are services typically bounded?

Service decomposition is breaking an application into distinct services. Services are typically bounded by business capability or Domain-Driven Design 'bounded contexts', keeping each service cohesive around one responsibility and loosely coupled from others.

What is a 'bounded context' in Domain-Driven Design?

An explicit boundary within which a particular domain model is defined and consistent. It maps well to microservice boundaries, ensuring each service owns its own model and data.

Contrast synchronous and asynchronous inter-service communication.

Synchronous (e.g., REST/gRPC request-response) has the caller wait for a reply, creating temporal coupling. Asynchronous (e.g., messaging/events) lets the sender continue without waiting, improving decoupling and resilience but adding eventual-consistency complexity.

What is gRPC and why is it used for inter-service communication?

gRPC is a high-performance RPC framework using HTTP/2 and Protocol Buffers for compact binary serialization. It offers low latency, strong typing via contracts, streaming, and is favored for internal service-to-service calls.

What is a service mesh?

A dedicated infrastructure layer that manages service-to-service communication, providing traffic management, load balancing, service discovery, observability, retries, and mutual TLS security — transparently, without changing application code.

In a service mesh, what is the sidecar proxy pattern?

A proxy (e.g., Envoy) deployed alongside each service instance that intercepts all inbound/outbound traffic, handling routing, security (mTLS), and telemetry. The collection of sidecars forms the mesh's data plane.

Distinguish the data plane and control plane in a service mesh.

The data plane consists of sidecar proxies that carry the actual service traffic. The control plane (e.g., Istio's istiod) configures and manages those proxies — distributing policies, certificates, and routing rules.

Name two popular service mesh implementations.

Istio (with Envoy proxies) and Linkerd; others include Consul Connect and AWS App Mesh.

What is an API Gateway and what functions does it provide?

A single entry point for client requests to backend services. It handles request routing, composition/aggregation, authentication/authorization, rate limiting/throttling, SSL termination, caching, and protocol translation.

How does an API Gateway differ from a service mesh?

An API Gateway manages north-south traffic (external clients to services) at the edge. A service mesh manages east-west traffic (internal service-to-service communication) within the cluster.

What is a message queue and what communication model does it enable?

A message queue stores messages sent by producers until consumers process them, enabling asynchronous, decoupled point-to-point communication. It buffers load, smooths spikes, and allows producers and consumers to operate independently.

In a point-to-point message queue, how many consumers process each message?

Exactly one consumer processes each message; once consumed and acknowledged, the message is removed from the queue (competing consumers share the workload).

What is the publish-subscribe (pub/sub) messaging model?

Publishers send messages to a topic without knowing subscribers; all interested subscribers receive a copy of each message. This one-to-many, fully decoupled model broadcasts events to multiple consumers.

Contrast message queues (point-to-point) with publish-subscribe.

In a queue, each message is delivered to a single consumer (work distribution). In pub/sub, each message is delivered to all subscribers of the topic (broadcast/fan-out).

What is event streaming and how does it differ from traditional message queues?

Event streaming captures events as an ordered, durable, replayable log that multiple consumers can read independently at their own offset, retaining events for a configured period. Traditional queues typically delete a message once consumed and don't support replay.

In Apache Kafka, what is a topic partition and why does it matter?

A partition is an ordered, append-only sub-log of a topic. Partitions enable parallelism (each consumed by one consumer in a group) and scalability; ordering is guaranteed only within a partition, not across the whole topic.

In Kafka, what is a consumer group and how does it relate to partitions?

A consumer group is a set of consumers sharing a subscription; Kafka assigns each partition to exactly one consumer in the group, so adding consumers (up to the partition count) increases parallel consumption. Multiple groups each get all messages independently.

Compare Apache Kafka and Amazon Kinesis for event streaming.

Both are distributed, partitioned, replayable streaming platforms. Kafka is open-source, self- or vendor-managed, uses topics/partitions with tunable retention. Kinesis is a fully managed AWS service using streams/shards with a default retention (24 hours, extendable up to 365 days), trading some control for reduced operational overhead.