Cassandra: Past, Present, and Future – History, Architecture, Features, and Use Cases

The talk, delivered by Chen Jiang (Alibaba Cloud Senior Expert) at a Cassandra Meetup and organized by DataFunTalk, introduced the theme "Cassandra's Past, Present, and Future" and provided a comprehensive overview of the database.

Origin : Cassandra was inspired by Google’s BigTable and Amazon’s Dynamo. From BigTable it adopted the LSM‑based single‑node engine concepts such as Column Families, Memtables, and SSTables, while Dynamo contributed the distributed design, cluster management, and fault‑tolerance techniques.

Milestones :

July 2008 – Facebook released Cassandra (c*).

2009 – Became an Apache incubator project.

2010 – Graduated to a top‑level Apache project.

2011 – 1.0 released with leveled compaction.

2013 – Introduced CAS and triggers.

2015 – 3.0 released.

2019 – 4.0 released.

Database Ranking : According to DB‑Engines, Cassandra consistently ranks first among wide‑column NoSQL databases, far ahead of HBase, with a popularity score above 100 compared to HBase’s ~50.

Current Feature Overview :

Peer‑to‑peer nodes enable easy horizontal scaling.

LSM engine provides high‑throughput writes.

High availability and fault tolerance via replication.

Tunable consistency levels.

CQL query language and JDBC‑like drivers.

Elastic data storage and straightforward data distribution.

Consistency Hash & Gossip : Cassandra uses a single‑hash partitioning scheme to map keys to token ranges, eliminating a master node. Nodes exchange metadata via a peer‑to‑peer gossip protocol, achieving eventual consistency while keeping metadata lightweight.

LSM Engine Details : Writes are first recorded in a Write‑Ahead‑Log, then stored in a Memtable. When the Memtable grows, it is flushed to an SSTable. Compaction strategies include size‑tiered, leveled, and time‑windowed compactions.

Adoption : Major companies use Cassandra, including Facebook (original creator), Apple (100k+ nodes), 360, Ele.me, Reddit, Discord, and many others for large‑scale workloads.

Value Propositions :

Always‑online with multi‑master replication and tunable consistency.

Linear scalability simplifies operations; adding a node automatically balances data.

Multi‑DC deployment reduces latency and provides geographic disaster recovery.

Rich client drivers for Python, C++, Go, Node.js, PHP, etc.

Strong performance: lower latency and higher throughput than HBase in many benchmarks.

Typical Use Cases :

Risk‑control systems (user profiles, fraud detection, order data).

Personalized recommendation engines (behavior analysis, real‑time processing).

Big‑data pipelines.

Social feeds (e.g., Instagram, Weibo‑like timelines).

Time‑series and IoT data ingestion with massive concurrent writers.

Future Roadmap (Cassandra 4.0‑alpha) :

Fix incremental repair bugs; recommend full repair with caution.

Replace custom node‑to‑node communication with Netty for higher efficiency.

Add built‑in time functions and arithmetic operators.

Expose SASI indexes and materialized views as experimental features.

Community Projects (NGCC 2019) :

Pluggable storage engine supporting RocksDB to reduce JVM GC pressure.

Sidecar – a one‑stop operations platform for bootstrapping, data movement, configuration upgrades, monitoring, backup/restore, and repair.

ScyllaDB improvements for more efficient data repair.

Next‑generation compaction strategies beyond leveled compaction.

Rocksandra : An Instagram‑driven effort to combine Cassandra with RocksDB, delivering lower GC overhead, reduced tail latency, and higher throughput.

Sidecar Details :

Handles bootstrap and data migration.

Integrates common fault‑tolerance and operational commands.

Provides configuration upgrades, monitoring, metrics, and enterprise‑grade backup/restore dashboards.

Offers repair and optimization utilities.

Overall, Cassandra remains a leading wide‑column NoSQL solution, distinguished by its master‑less architecture, linear scalability, multi‑DC capabilities, extensive language drivers, and a vibrant community driving continuous innovation.