OmScaler is a partner program that makes larger OmDB cluster with multiple OmDB backends.
- Extends OmDB to larger scale.
- Sits in front of OmDBs.
- Maps multiple OmDB nodes into linear namespace. (No Consistent-Hash rings)
- Provides data availability by data replication,
- Provides Quorum-based data integrity.
The problems of Consistent-Hash based NoSQL like Cassandra.
- It doens't scale well due to the nature of required re-balancing when number of nodes changes. In large scale, server always dies.
- Consistent-hash model is hard to handle hot-spot due to the nature of the algorithm.
- Performance lag grows as cluster grows.
- If serious sever failure happens, it's likely to result global service failure.
- Separate 2nd index is needed to iterate keys.
What about OmScaler's linear partitioning?
- OmScaler uses partition map that maps range of namespace to each set of OmDBs.
- Yeah, there would be some overhead on partition management.
- But much lesser data re-balancing and linear performance increase as cluster grows. So it can scale out.
- Isolated failure. Even if serious of server failure happens, only that range of data gets out of service.
- No index is needed for key iteration.
Note) OmScaler is not publicly available, please contact the author.