Hazelcast Releases 3.8 – The Fastest Open Source In-Memory Data Grid

Print Friendly, PDF & Email

Hazelcast, a leading open source in-memory data grid (IMDG) with hundreds of thousands of installed clusters and over 17 million server starts per month, announced the general availability of Hazelcast IMDG 3.8.

A major theme of Hazelcast IMDG 3.8 is better operational experience for systems requiring maximum uptime. The number of times the cluster must be shut down for maintenance are dramatically reduced. Advanced data management capabilities are added for managing persistence and multi-data center deployments. Hazelcast is being used in more and more five nines applications where taking maintenance windows is difficult.

Available in 3.8, Rolling Member Upgrades enable users to upgrade cluster Members without service interruption. This adds to the client rolling upgrade feature introduced in 3.6 a year ago. Combined with User Code Deployment this dramatically reduces the need to ever restart a production Hazelcast cluster.

In Hot Restart Store we add the ability to restart a cluster after a lights out event with partial failures such as servers that cannot be restarted. We also allow backed up data to be copied to new machines and used to restart a new cluster with data intact. Plus we add the ability to take a snapshot of the Hot Restart Store in a running cluster for periodic backups that can be saved offline. These features add up to much higher persistence.

WAN Replication is also enhanced with new Dynamic WAN Synchronization allowing users to copy one cluster’s data to another without service interruption. The sync process can be started inside the WAN Sync interface of Management Center. Therefore, at any time, users can create a new WAN replication destination and a snapshot of their current cluster using sync ability. The use case is to allow a new data center to be brought online without service interruption.

User Code Deployment allows the user code that needs to be deployed to Hazelcast to allow in-situ processing, for example EntryProcessors, to be done dynamically from a Member or Lite Member, thus removing the need to copy jars to each Member and restart.

Hazelcast has a large community which drive feature development. Responding to this 3.8 includes numerous enhancements with the highlights being:

  • Fast Aggregations on top of query execute in a fraction of the time of the old Aggregations capability. All of the standard aggregations like sum and count have been built-in. This function has been moved from MapReduce to more conveniently sit on our Query API. Performance tests show it 3x faster than competitors.
  • Query Projection have also been added so that specific fields of an entry value rather than the whole value can be returned. This is very important for Query performance and cuts the network and serialization overhead.
  • Scheduled Executor Service, a feature that enables developers to schedule tasks at any time, or conduct repetitive scheduling at fixed intervals in a cluster.
  • Near Cache Preload – Client Near Caches can be configured to preload entries from the cluster when the client application starts, so that near caches are hot before use. Near caches can speed up performance more than 10x.
  • Node.js client improved with the most frequently used Hazelcast data structures. These include Distributed Map with support of Predicates and Entry Processors, MultiMap, Set, List, Distributed Locks and Queue.
  • Continuous Query Cache has been open sourced in 3.8. With this, results of a query are always ready and local to the application with zero latency.
  • HyperLogLog is a new probabilistic data structure used to “estimate” cardinalities of unique elements on huge sets. Common use cases include calculating real time unique site visitor metrics based on IP or user and measuring campaign performance (impressions, clicks, etc) in advertising.
  • Ring Buffer Store – a storage mechanism for Ring Buffer, which when enabled allows the reading of items which are no longer in the Ring Buffer.
  • Increase split-brain protection to cover the Queue and Lock data structures in addition to the existing Map and Cache. This prevents writes and/or reads from occurring with sub-clusters which have formed after a network partition.

There are many important and useful enhancements in this release which elevates Hazelcast IMDG to a new level,” said Greg Luck, CEO of Hazelcast. “This couldn’t be done without community support, their contributions are vital if Hazelcast is to continue to move forward and offer a solution that is custom-made for specific industry requirements. Many businesses must now offer services in real time, with the enhancements and improvements offered in 3.8, Hazelcast users will be able to build high performance applications safe in the knowledge that the end user expectation will be met regardless of load.”


Sign up for the free insideBIGDATA newsletter.

Speak Your Mind