Red Hat
Sep 16, 2015
by Radim Vansa
Infinispan local caches have several features that make it more than just a map - expiration and eviction, listeners, statistics, transactions, cache stores and so forth. However, this comes at a price - due to all the hooks and object allocation, plain ConcurrentHashMap is faster than local cache.

There are applications that need something in between - great performance but some of those features, too. In our case, the motivation were internal caches in Hibernate Second Level Cache. Therefore Infinispan 8.0.1.Final brings the simple cache - alternative implementation of the AdvancedCache interface optimized for maximum performance when you need just the basics.

The table below shows which features are available in simple cache:

Basic map-like API
Cache listeners (non-clustered)
JMX access
Invocation batching
Persistence (cache stores and loader)
Map Reduce Framework
Distributed Executors Framework
Custom interceptors
Indexing (querying)
Compatibility (embedded/server)
Store as binary

Configuring simple cache is as simple as adding one attribute to the XML configuration:

While configuration schema allows to set up the unsupported features, doing so results in an exception when the cache is created.

You can also configure simple cache programmatically:

So, what kind of performance improvement can you expect? We had run basic (single-threaded) benchmark using JMH and this is what we got:

Implementationget() (operations/s)put() (operations/s)
Simple cache86,969,897±738,93514,044,642±14,280
Local cache17,280,018±361,9102,267,850±44,814

This gives us about 5✕ speedup for reads and 6✕ for writes. Your mileage may vary, but it's certain that simple cache provides substantial performance benefits.

So, if your use-case allows it, try out simple cache and let us know. It's as simple as one configuration attribute!

Original Post