Cloud Computing Core Cloud Infrastructure Services Syllabus

What does '11 nines' (99.999999999%) of durability imply in practical terms?

It means an expected annual loss of about one object per 100 billion objects—i.e., if you store 10 million objects you would expect to lose one object roughly once every 10,000 years. Achieved by redundant, replicated storage.

Contrast synchronous and asynchronous replication.

Synchronous: a write is acknowledged only after all replicas confirm, giving zero data loss (RPO≈0) at the cost of higher latency and tighter distance limits. Asynchronous: primary acks immediately and propagates later, giving lower latency but possible data loss on failure.

What are RPO and RTO in the context of replication and disaster recovery?

RPO (Recovery Point Objective): maximum acceptable data loss measured in time (how far back the last usable copy is). RTO (Recovery Time Objective): maximum acceptable downtime to restore service. Lower RPO/RTO cost more and drive replication strategy.

How does erasure coding provide durability more efficiently than full replication?

Erasure coding splits data into $k$ data fragments plus $m$ parity fragments, so any $k$ of the $k+m$ fragments reconstruct the data—tolerating $m$ failures with storage overhead $\frac{k+m}{k}$, far less than the 3x overhead of triple replication.

What is a Virtual Private Cloud (VPC)?

A logically isolated, software-defined virtual network within a public cloud where you control your own IP address range (CIDR block), subnets, route tables, gateways, and security controls—giving a private, customizable network for your resources.

Distinguish public and private subnets within a VPC.

A public subnet has a route to an Internet Gateway, so its resources can have public IPs and reach/receive from the internet. A private subnet has no direct internet route; its instances reach the internet outbound only via a NAT gateway and are not directly reachable inbound.

Compare security groups and network ACLs (NACLs) in a VPC.

Security groups are stateful, operate at the instance/interface level, and support allow rules only (return traffic auto-allowed). NACLs are stateless, operate at the subnet level, support both allow and deny rules, and evaluate return traffic explicitly by rule order.

What does CIDR notation like 10.0.0.0/16 specify, and how many addresses does it contain?

CIDR gives a base IP plus a prefix length; /16 fixes the first 16 bits as network, leaving $32-16=16$ host bits, so $2^{16}=65{,}536$ total addresses. Larger prefix number = smaller subnet (e.g., /24 = $2^{8}=256$ addresses).

What is a load balancer and what core problem does it solve?

A load balancer distributes incoming traffic across multiple backend targets to prevent any single server from being overwhelmed, improving availability, fault tolerance (routing around unhealthy targets), and horizontal scalability.

Contrast Layer 4 and Layer 7 load balancing.

Layer 4 (transport) load balancers route by IP/TCP/UDP without inspecting content—fast, protocol-agnostic. Layer 7 (application) load balancers inspect HTTP(S) content, enabling routing by URL path, host header, or cookies, plus TLS termination and content-based rules.

Name common load-balancing algorithms and what each does.

Round robin (rotate through servers), Least connections (send to server with fewest active connections), Weighted (bias by server capacity), IP hash / sticky sessions (map a client consistently to one server), and Least response time.

What is DNS and what does it fundamentally do?

The Domain Name System is a distributed, hierarchical directory that resolves human-readable domain names (e.g., example.com) into IP addresses, plus other records, acting as the internet's phone book so clients can locate servers.

Match common DNS record types to their function: A, AAAA, CNAME, MX, TXT.

A: maps a name to an IPv4 address. AAAA: maps to an IPv6 address. CNAME: aliases one name to another canonical name. MX: designates mail servers for a domain. TXT: holds arbitrary text (SPF, domain verification, DKIM).

What is a Content Delivery Network (CDN) and how does it improve performance?

A CDN is a geographically distributed network of edge/cache servers that store copies of content close to users. Requests are served from the nearest edge, reducing latency, offloading the origin, absorbing traffic spikes, and improving availability.

Differentiate a CDN cache hit from a cache miss and the role of TTL.

A cache hit serves content directly from the edge (fast, no origin call). A miss means the edge lacks/expired content and fetches from the origin, then caches it. TTL (time-to-live) sets how long an edge keeps content before revalidating with the origin.

Compare a site-to-site VPN and a Direct Connect / dedicated interconnect for cloud connectivity.

VPN: encrypted tunnel over the public internet—quick, cheap, but variable latency/bandwidth. Direct Connect/Interconnect: a private physical link between on-premises and the cloud—consistent low latency, higher/dedicated bandwidth, better security, but costlier and slower to provision.

What is VPC peering and a key limitation of it?

VPC peering privately connects two VPCs so resources communicate using private IPs as if on one network. Key limitation: it is non-transitive—if A peers B and B peers C, A cannot reach C through B; each pair needs its own peering, and CIDR ranges must not overlap.

What problem do transit gateways / network hubs solve that peering does not?

They provide a central hub to which many VPCs and on-premises networks connect once, enabling transitive, hub-and-spoke routing among all of them—avoiding the $\frac{n(n-1)}{2}$ mesh of individual peerings needed to fully connect $n$ VPCs.

What is an Internet Gateway and what does it enable in a VPC?

An Internet Gateway is a horizontally scaled VPC component that allows bidirectional communication between the VPC and the public internet. It performs network address translation for instances with public IPs, enabling them to send and receive internet traffic.

What is a NAT Gateway and why is it used?

A NAT (Network Address Translation) Gateway lets instances in a private subnet initiate outbound internet connections (e.g., for updates/patches) while preventing unsolicited inbound connections from the internet, preserving the privacy of private-subnet resources.

Compare an Internet Gateway and a NAT Gateway on directionality.

Internet Gateway: allows both inbound and outbound internet traffic for public-subnet resources with public IPs. NAT Gateway: allows only outbound-initiated traffic (and its return) for private-subnet resources—no inbound internet-initiated connections.

What defines a relational database (RDBMS) and its data model?

An RDBMS stores structured data in tables (rows and columns) with a predefined schema, related via primary/foreign keys. It is queried with SQL and enforces relationships and constraints, ensuring data integrity through normalization.

What do the ACID properties guarantee in a relational database?

Atomicity (a transaction fully completes or fully rolls back), Consistency (transactions move the DB between valid states honoring constraints), Isolation (concurrent transactions don't interfere), and Durability (committed changes survive crashes).

What is a read replica and how does it help scale a relational database?

A read replica is an asynchronously updated copy of the primary database that serves read-only queries, offloading read traffic from the primary. It scales read-heavy workloads horizontally and can be promoted to primary for disaster recovery.

What is a NoSQL database and what are its four main types?

NoSQL databases store non-relational, often schema-flexible data optimized for scale and specific access patterns. Four types: key-value (e.g., simple lookups), document (JSON-like documents), wide-column (column families), and graph (nodes and edges/relationships).

State the CAP theorem and its implication for distributed databases.

CAP theorem: in the presence of a network Partition, a distributed system can guarantee at most one of Consistency or Availability (not both). Since partitions are unavoidable, systems choose to be CP (consistent) or AP (available) during a partition.

Contrast strong consistency and eventual consistency.

Strong consistency: every read returns the most recent write immediately—simpler correctness, higher latency. Eventual consistency: reads may return stale data briefly, but all replicas converge to the same value over time—higher availability and lower latency, common in AP NoSQL systems.

What is in-memory caching and why is it fast? Give example uses.

In-memory caching stores frequently accessed data in RAM (e.g., Redis, Memcached) instead of disk, so reads are sub-millisecond because RAM access is orders of magnitude faster than disk. Used for session stores, query-result caching, leaderboards, and rate limiting.

Explain cache eviction and two common policies: LRU and TTL.

When a cache is full, eviction removes entries to make room. LRU (Least Recently Used) evicts the entry unused for the longest time, keeping hot data. TTL (time-to-live) expires entries after a fixed duration to bound staleness regardless of use.

How is a data warehouse different from an OLTP relational database?

A data warehouse is optimized for OLAP—analytical queries scanning huge historical datasets—often using columnar storage and massively parallel processing. OLTP databases are optimized for many small, fast read/write transactions (row-oriented). Warehouses prioritize read/aggregate throughput over transactional writes.

Why does columnar storage speed up analytical (OLAP) queries?

Columnar storage keeps each column's values together, so analytical queries that aggregate a few columns read only those columns (less I/O), and similar values compress far better. Row storage would force reading entire rows even when only a few columns are needed.