Datastax recently announced the release of Cassandra 1.2.0 and listed some of it's key improvements. From the original post:
Key improvements include:
- Virtual nodes, which improve the granularity of capacity increases and dramatically improve repair and rebuild times in larger clusters. See also this post on upgrading an existing cluster to vnodes.
- CQL3 improvements, notably the addition of collection types, queryable system information, and a CQL-native protocol.
- Request tracing is available to both CQL and classic Thrift requests, and can also be managed programatically.
- Atomic batches address the possibility of mid-batch coordinator failure.
- Configurable policies for disk failure
- Last but not least, many performance improvements to memory usage, column indexes, compaction, streaming, startup time, and more.