ScyllaDB is a high-performance C++ Cassandra look-alike that is quickly shaping up to be an interesting alternative. Currently DTCS support is still weak, causing RB tests to hang on table creation. Improvements for it are scheduled in Q1. Encryption is also still missing.
Lets revisit Scylla once DTCS and encryption have landed. http://www.scylladb.com/technology/status/ is a good place to track progress.
Potential benefits
- About an order of magnitude more / smaller instances (one instance per core)
- Higher data locality reduces SSTables / read independent of compaction strategy.
- Reduced LSM write amplification might enable the use of LCS, which should address range query (wide row), and tombstone GC issues seen with other compaction strategies.
- vnode tokens implicitly shared across instances: Reduces repair write amplification by number of instances, without introducing storage imbalance.
- Automatic instance management: No manual bootstrapping, restarting etc of individual instances.
- More efficient implementation using C++ instead of Java. No GC.
Potential issues
- New, less mature software.
- Less complete feature set at this point:
- No incremental repairs.
- No Cassandra 3.x features like materalized views.
- Cost of migration.