Distributed Systems

Vertical Scaling (Scale Up): Add more power (CPU, RAM, SSD) to a single machine. Upgrading from 8-core → 32-core server.
- Hardware limit — one machine can only get so big
Horizontal Scaling (Scale Out): Add more machines/nodes to distribute data and load.
- Software complexity — requires data distribution & coordination

To achieve this, they all use:

Cluster: A group of servers (nodes) working together as one logical system. [Kafka, Spark, Cassandra, Elasticsearch]
Node: A single machine or instance in the cluster. [Kafka, Spark, Cassandra, Elasticsearch]
Partitioning: Splitting data/work into smaller pieces so each node handles a subset. [Kafka (topics), Cassandra (tables), Elasticsearch (indices), Spark (RDDs)]
Replication: Making copies of data to ensure fault tolerance. [Kafka, Cassandra, Elasticsearch]
Sharding: Usually synonymous with partitioning (especially for databases). Sometimes refers to a physical division of data across machines. [Cassandra, Elasticsearch]
Leader/Follower: One node manages writes (leader), others replicate it (followers). [Kafka, Cassandra, Elasticsearch]

Partitioning (or sharding): Splitting the dataset into smaller, non-overlapping parts (partitions) that can be distributed across multiple nodes for scalability and parallelism.
Replication: Copying each partition to multiple nodes so that if one node fails, data is still available from another node — ensures reliability and fault tolerance.

👉 Partitioning → Split across nodes (scale) → Horizontal Scaling

👉 Replication → Copy across nodes (protect) → Fault Tolerance

Each partition (P0, P1, P2) contains a distinct subset of data.
- → Enables parallelism and scalability.
Each partition is replicated to another node.
- → Ensures fault tolerance.
If Node B fails: Node A and C still contain replicas for all data partitions.

Google Sites

Report abuse