Java UUID generators compared

Posted 2007-04-25 in Java by Johann.

UUIDs (or GUIDs) are unique identifiers that are frequently used in programming. In this entry, I will compare three generators of UUIDs written in the Java programming language.

Possible uses for UUIDs

UUIDs have a variety of uses, including

• Temporary file names.
• Unique identifiers for website visitors.
• Transaction IDs.
• Primary keys, replacing database sequences.

Generally, whenever unique values across different machines are needed, UUIDs are a good choice.

The contestants

The following packages were compared:

• UUID – my own implementation.
• Java UUID Generator.
• java.util.UUID, available since JDK 5.

I did not compare the following packages:

• The implementation found in Jini.
• Commons Id – there is no release as of now.

The features

I compared the following features:

• Ease of use: How many lines of code does it take to get a UUID?
• Versions: The UUID versions supported (1–5).
• Speed: The performance of generating UUIDs.
• EJB: Whether the generated UUIDs can be used as primary keys in enterprise java beans.
• CORBA: Whether the generated UUIDs can be used in CORBA.
• Parse: Whether parsing UUIDs is supported. Usually, this is done from java.lang.Strings, although JUG also supports parsing from byte arrays.

The results

Name	Ease of use	Versions	Speed	EJB	CORBA	Parse
++: very good, +: good, o: acceptable, -: bad, --: very bad
UUID	++	1	++	Yes	Yes	String, CharSequence
JUG	o	1, 3–5	+/-/--	No	No	String, byte[]
JDK 5	++	3, 4	-/--	No	No	String

The benchmark

The performance of all generators was compared. The test computer was an Intel Pentium M 1.73 GHz computer with 1.5 GB of RAM. Load on the computer was low during the tests.

The following Java Virtual Machines were used:

• Sun JDK 1.5.0_05
• IBM 1.3.1
• Sun 1.4.2_09
• BEA JRockit 1.4.2_08
• BEA JRockit 1.5.0_08
• Sun JDK 6

All Java Virtual Machines were started with -Xms256M -Xmx256M to force a fixed VM size.

The data

The following table contains the timings in milliseconds for the creation of one million UUIDs with the various implementations.

VM	UUID	JUG, time	JUG, name	JUG, random	JDK, random	JDK, name
Sun 1.5.0_05	1315	3424	25518	25946	28565	35253
IBM 1.3.1	2359	6796	81134	46584
Sun 1.4.2_09	2800	7053	91569	64956
JRockit 1.4.2_08	2556	5859	63890	47500
JRockit 1.5.0_08	1431	3756	25653	35093	35797	32253
Sun JDK 6	2303	5703	35968	41100	41553	48859

My own implementation is the fastest by a factor of at least two. With a modern computer, it is possible to generate more than 500.000 UUIDs per second.

The good

All of the implementations I tested generated unique UUIDs, even when used multi-threaded.

The bad

Some of the implementations are slower than others. It might not be possible to use some implementations in environments like EJB containers.

The MAC address issue

The computer’s MAC address must be encoded into a version 1 UUID. Because there is no official way in Java to do this, JUG comes with a native library (compiled for Windows x86, FreeBSD, Linux x86, MacOSX PPC, SPARC). My own version does this without native code (tested on Windows, Linux, MacOSX, BSDs, Solaris, HP-UX) and falls back to random data if the MAC address cannot be read.

Security aspects must be considered when version 1 UUIDs that contain the MAC address are made public.

The summary

If you need speed and/or support for EJBs or CORBA, UUID is the best way. If security is of ultimate importance to you, it cannot be used. In that case, JUG in the time-based mode is a good choice.

2 comments