Object Pools revisited

Up to Java 5.0 using object pools could significantly improve performance. However from Java 5.0, the cost of using naive object pools was often worse (and more complicated) than just allocating new objects all the time. However, if you work in a low latency environment and you want to keep allocations low, is there a way to make an object pool light weight enough to make sense.

A common use of a low level object is in the creation of Strings esp as input. Whether you have a service reading data from the network, or loading data from a database or file, you are likely to use a lot of Strings. While there are tricks to avoid using Strings, most of the alternatives lead to unnatural Java code which is ugly and harder to maintain. It would be much better if there was an efficient way to reuse Strings (assuming a high rate of duplication)

This test assume we have some data in native memory (from a TCP socket or a file), and we have the option to avoid copying the data into a byte[] before creating a String.

For different lengths of bytes (1, 2, 4, 8, 16, 32) we use JMH to benchmark how long it takes to provide a matching String.

"Object Pool" uses an object pool for the bytes converted into a String.

"new String US-ASCII" copies the data to a byte[] and uses US-ASCII encoding to create the String.

"new String hibyte=0" copies the data to a byte[] and uses a deprecated constructor to set the hi byte to 0 create the String.

The Source code is available here ObjectPoolMain.java

The average time to use the Object Pool was around 9 nano-seconds for short strings of up to 8 characters. About half the saving comes from avoiding the need to copy the data to a byte[] first. Also avoiding the use of even a trivial Charset character encoding also saved a significant amount.

In this case the main reason is to avoid creating objects. If the String can be reused, it can dramatically reduce the garbage produced. Most of the time, only new Strings take space, rather than every String your use and this can be a 90% reduction in memory use and garbage production.

For some use cases, an Object Pool is fast enough, possibly faster when it is has been optimised to minimise copies and supporting functions like character encoding.