ScyllaDB is an increasingly popular alternative NoSQL database to Apache Cassandra. Imagine all the goodness of Cassandra — same highly available ring architecture, same SSTable data format, same Apache Cassandra Query Language (CQL) — but without the Java overhead.
With ScyllaDB you will achieve higher performance at scale using dramatically fewer nodes, with far less administration, and lower infrastructure costs. You can even switch from Cassandra to ScyllaDB without interruption and commonly with no code changes. ScyllaDB is available as a fully managed service, an enterprise offering, and open source.
ScyllaDB was built for performance at scale, delivering lower latency and and higher throughput through its close-to-the-metal design.
Moving to ScyllaDB commonly reduces clusters to 1/10th the size of Cassandra, delivering significant cost savings and greater ROI.
Cassandra needs a lot of tuning & upkeep. ScyllaDB automatically tunes its own IO & CPU performance & simplifies operations.
ScyllaDB scales out more efficiently and also scales up to fully utilize higher capacity nodes.
ScyllaDB Migrator lets you move your data from your existing Cassandra clusters into ScyllaDB with zero hassle or downtime.
ScyllaDB provides features Cassandra cannot match, like production-ready materialized views, global secondary indexes & more.
Comcast Xfinity improved user experience and shrank their cluster from 962 nodes with Cassandra to 78 nodes on ScyllaDB, reducing administrative overhead over 90%.
Rakuten’s billion-item catalog 6x better performance while shrinking their cluster 70% moving from Cassandra to ScyllaDB. They also eliminated unpredictable latencies.
Fanatics replaced 55 nodes of Cassandra with just 6 nodes of ScyllaDB. Discover why the brand that powers all NFL, MLB, NBA, and NHL shops switched to ScyllaDB.
Expedia, the travel booking giant, reduced P99 response times from >80ms to <5ms after migrating to ScyllaDB NoSQL. They also got 3x the throughput while saving 35%.
Learn why and how Discord’s persistence team completed their most ambitious migration yet: moving their massive set of trillions of messages from Cassandra to ScyllaDB.
GE Predix reduced administrative burden in their IIoT infrastructure to meet their SLAs, and shrank their Cassandra cluster of 75 nodes to just 15 nodes of ScyllaDB.
ScyllaDB
Cassandra
Both NoSQL databases use a SQL-like query language; users do not need to alter queries moving to ScyllaDB from Cassandra.
ScyllaDB can take advantage of any CQL driver used by Apache Cassandra. These are available in many popular programming languages. Users can keep the exact same drivers if they wish.
| ScyllaDB offers faster shard-aware drivers that provide superior performance in conjunction with ScyllaDB’s shard–per-core design. These drivers can add another 10% or more performance boost. |
Allows users to specify various levels of consistency to tailor operations to their use cases. Consistency is set on an operation-by-operation basis.
| Scale out to any number of nodes in a cluster. Since ScyllaDB is more efficient it can run the same workload on fewer nodes. |
| ScyllaDB scales up on larger, denser nodes with any number of CPUs, and any amount of RAM or storage. Cassandra is limited to 2 terabytes of data on a single node. |
| ScyllaDB accomplishes topology changes far faster than Cassandra, significantly reducing the time to resize and rebalance clusters. |
ScyllaDB maintains smooth, single-digit millisecond P99 latencies and high throughput measured in millions of OPS. ScyllaDB uses less infrastructure to run the same workload as Cassandra with better outcomes.
| ScyllaDB allows users to set prioritization for specific workloads, minimizing chances they have to be retried or dropped. Unique to ScyllaDB Enterprise and ScyllaDB Cloud. |
Automatically balances and redistributes data across all the nodes in a cluster.
ScyllaDB and Cassandra both use active-active clustering. Every node accepts reads and writes.
Both offer cross-datacenter clustering and automatic multi-datacenter data replication.
Remains highly available for operations when facing node outages and network partitions.
ScyllaDB’s CPU and I/O schedulers make adaptive changes to database behavior for optimum performance in real time. Cassandra requires complex manual JVM tuning which cannot reflect changing real-world conditions.
ScyllaDB, written in C++, does not suffer from Cassandra’s “stop-the-world” pauses caused by Java Garbage Collection (GC).
| ScyllaDB is NUMA-aware so your data in memory can be closely aligned to its assigned CPU core. The JVM Cassandra uses may or may not be NUMA-aware. |
| Both databases use SSTables — a fast, immutable Log Structured Merge (LSM) tree based storage format. |
| Both ScyllaDB & Cassandra support multiple compaction strategies to remove obsolete data and minimize storage on disk. |
ScyllaDB’s incremental compaction strategy allows greater storage utilization. Unique to ScyllaDB Enterprise and ScyllaDB Cloud.
ScyllaDB’s I/O scheduler prioritizes foreground operations (reads and writes) over background compaction tasks, lowering impact during peak workloads.
Materialized views allow unique organization of information. Only ScyllaDB’s materialized views are production-ready; they are still experimental in Cassandra.
Allow users to organize information using non-partition key columns, but are limited to data stored in a specific node.
Global secondary indexes span all nodes in a cluster; they are more powerful but come at a cost in performance. Only ScyllaDB supports global secondary indexes
Lightweight Transactions, also called compare-and-set (CAS) operations, do a read-before-write to ensure strict linearizability (ordering) of operations. They use the Paxos consensus protocol to provide ACID-like consistency on a single partition basis. ScyllaDB’s LWT implementation is more efficient than Cassandra’s.
ScyllaDB offers CDC as standard CQL-queryable tables, making it easily consumable by applications and event streaming systems like Apache Kafka. Cassandra’s CDC implementation does not use the CQL interface and is not as elegant or efficient.
ScyllaDB uses a highly-efficient row-based data cache. Cassandra uses multiple layers of caching, including the OS-level cache, which is inelegant and inefficient.
ScyllaDB allows queries to bypass the cache entirely on a per-operation basis, which is useful for range and full table scans.
Apache® and Apache Cassandra® are either registered trademarks or trademarks of the Apache Software Foundation in the United States and/or other countries. Amazon DynamoDB® and Dynamo Accelerator® are trademarks of Amazon.com, Inc. No endorsements by The Apache Software Foundation or Amazon.com, Inc. are implied by the use of these marks.
