benchmark-it

Performance measurements for common code samples

Adding new benchmark

Find the following official examples of using JMH: org/openjdk/jmh/samples/

Also you can take a look at my own measurements. The simplest is StringConcat.java. It demonstrates why we actually need classes like StringBuilder.

Running benchmarks

Then you can use run-benchmarks.sh script to execute benchmarks (You need to have maven been installed):

./run-benchmarks.sh "com.github.abrasha.strings.StringConcat"

Existing benchmarks:

Cloning v. creating new objects:

Benchmark                       Mode  Cnt   Score   Error  Units
CloneVsNewWithSetters.viaClone  avgt   10  10.041 ± 0.059  ns/op
CloneVsNewWithSetters.viaNew    avgt   10   7.617 ± 0.113  ns/op

Reflection v. direct constructor call:

Benchmark                                            Mode  Cnt   Score     Error   Units
ConstructorVsReflection.createPersonWithConstructor  avgt    9   5.323   ± 0.233   ns/op
ConstructorVsReflection.createPersonWithReflection   avgt    9  12.911   ± 1.476   ns/op

HashMap v. LinkedHashMap v. Hashtable as local property storage (10 sequential insertions):

Benchmark                             Mode  Cnt    Score    Error  Units
HashMapVSLinkedHashMap.hashMap        avgt    9  101.183 ±  7.116  ns/op
HashMapVSLinkedHashMap.hashtable      avgt    9  242.710 ± 88.953  ns/op
HashMapVSLinkedHashMap.linkedHashMap  avgt    9  150.633 ± 15.713  ns/op

String concatenation v. StringBuilder v. StringBuffer:

Benchmark                             Mode  Cnt          Score           Error  Units
StringConcat.concatWithStringBuffer   avgt   10     209801.407 ±      2555.216  ns/op
StringConcat.concatWithStringBuilder  avgt   10     206855.786 ±      1913.781  ns/op
StringConcat.concatWithStringConcat   avgt   10  259190789.987 ± 262685019.287  ns/op

Iterating collection:

Benchmark                             Mode  Cnt    Score    Error  Units
HashMapVSLinkedHashMap.hashMap        avgt    9  101.183 ±  7.116  ns/op
HashMapVSLinkedHashMap.hashtable      avgt    9  242.710 ± 88.953  ns/op
HashMapVSLinkedHashMap.linkedHashMap  avgt    9  150.633 ± 15.713  ns/op

Should we do manual boxing / unboxing?

Benchmark                          Mode   Cnt  Score   Error  Units
AutoboxingVsManual.autoUnboxing    avgt    9   2.338 ± 0.047  ns/op
AutoboxingVsManual.manualUnboxing  avgt    9   2.316 ± 0.088  ns/op

How we should parse int from String?

Benchmark                      Mode   Cnt    Score   Error  Units
IntegerParsing.integerParse    avgt    9    23.314 ± 1.402  ns/op
IntegerParsing.integerValueOf  avgt    9    25.892 ± 0.758  ns/op

Integer.valueOf(int) for cached Integers.

Benchmark                         Mode  Cnt  Score   Error  Units
IntegerAutoBoxCache.getCached     avgt    9  2.869 ± 0.094  ns/op
IntegerAutoBoxCache.getNonCached  avgt    9  3.997 ± 0.107  ns/op

Proxy v. Non-Proxy class:

Benchmark                 Mode   Cnt  Score   Error  Units
ProxyVsNonProxy.nonProxy  avgt    9  72.668 ± 4.351  ns/op
ProxyVsNonProxy.proxy     avgt    9  71.439 ± 1.857  ns/op

String#startsWith(xxx) v. Pattern.compile("xxx.*"):

Benchmark                                    Mode   Cnt  Score   Error  Units
RegexVsStartsWith.startsWith                 avgt    9   5.404 ± 0.188  ns/op
RegexVsStartsWith.usingRegex                 avgt    9  88.533 ± 6.107  ns/op
RegexVsStartsWith.usingRegexWithStartAndEnd  avgt    9  48.726 ± 1.669  ns/op

String#matches(xxx) v. Pattern.compile(xxx).matcher(yyy).matches()

Benchmark                                  Mode   Cnt   Score    Error  Units
RegexMatcherVsStringMatches.regexMatcher   avgt    9   86.363 ±  3.194  ns/op
RegexMatcherVsStringMatches.stringMatcher  avgt    9  304.525 ± 17.621  ns/op

Creating new Cipher v. using ThreadLocal<Cipher>

Benchmark                       Mode  Cnt     Score     Error  Units
CryptoInit.withAvailableCipher  avgt   30  1795.563 ±  37.173  ns/op
CryptoInit.withNewCipher        avgt   30  6979.507 ± 144.121  ns/op