* The data is always stored in-memory (RAM) of the servers * Multiple copies are stored in multiple machines for automatic data recovery in case of single or multiple server failures * The data model is object-oriented and non-relational * Servers can be dynamically added or removed to increase the amount of CPU and RAM * The data can be persisted from Hazelcast to a relational or NoSQL database * A Java Map API accesses the distributed key-value store

* Local, Invalidation, Replicated and Distributed cache modes * Multiple discovery algorithms for elastic node provisioning * Partition handling detects split brain scenarios and reduces availability to provide consistency in such situations * Store immortal and mortal entries in the same cache * Entry mortality defined by lifespan or maximum idle time * Define the maximum number of entries to be kept in memory * Define the maximum size of entries to be kept in memory * Concurrency control based on Optimistic and pessimistic locks and total order * Support for XA and synchronization models * Synchronous and asynchronous indexing engine * Lucene-based indexing powered by Hibernate Search * Clustering authentication and authorization integrated with the Security Realms in the server. * Distributed streams based on java.util.stream * High-performance multi-threaded Map/Reduce * Distributed executor