聊聊G1 GC的String Deduplication

序

本文主要研究一下G1 GC的String Deduplication

-XX:+UseStringDeduplication

• jdk8u20给G1 GC带来了String Deduplication特性来将相同的字符串指向同一份数据，来减少重复字符串的内存开销
• 该特性默认是关闭的，可以使用-XX:+UseStringDeduplication来开启(前提是使用-XX:+UseG1GC)
• 具体的实现大致是JVM会记录char[]的weak reference及hash value，当找到一个hash code相同的String时，就会挨个char进行比较，当所有都match，那么其中一个String就会修改指针指向另一个String的char[]，这样前者的char[]就可以被回收

实例

实验代码

    @Test
    public void testG1StringDeduplication() throws InterruptedException {
        List<String> data = IntStream.rangeClosed(1,10000000)
                .mapToObj(i -> "number is " + ( i % 2 == 0 ? "odd" : "even"))
                .collect(Collectors.toList());
        System.gc();
        long bytes = RamUsageEstimator.sizeOfAll(data);
        System.out.println("string list size in MB:" + bytes*1.0/1024/1024);
        System.out.println("used heap size in MB:" + ManagementFactory.getMemoryMXBean().getHeapMemoryUsage().getUsed()*1.0/1024/1024);
        System.out.println("used non heap size in MB:" + ManagementFactory.getMemoryMXBean().getNonHeapMemoryUsage().getUsed()*1.0/1024/1024);

    }

关闭StringDeduplication

-XX:+UseG1GC -XX:-UseStringDeduplication

输出如下：

string list size in MB:586.8727111816406
used heap size in MB:831.772346496582
used non heap size in MB:6.448394775390625

• 整个jvm heap占用了约831MB，其中string list占用了约586MB

开启StringDeduplication

-XX:+UseG1GC -XX:+UseStringDeduplication

输出如下：

string list size in MB:296.83294677734375
used heap size in MB:645.0970153808594
used non heap size in MB:6.376350402832031

• 整个jvm heap占用了约645MB，其中string list占用了约296MB

小结

• jdk8u20给G1 GC带来了String Deduplication特性来将相同的字符串指向同一份数据，来减少重复字符串的内存开销
• 该特性默认是关闭的，可以使用-XX:+UseStringDeduplication来开启(前提是使用-XX:+UseG1GC)
• 在有大量重复string的前提下，使用G1 GC开启String Deduplication确实能够节省一定的内存，可以节约20%左右的内存，不过这个是理想的情况，因为普通应用里头的string重复的可能不多

doc

• Java 8 String deduplication vs. String.intern()
• JEP 192: String Deduplication in G1
• String Deduplication – A new feature in Java 8 Update 20
• Java 8 Update 20: String Deduplication
• Duplicate Strings: How to Get Rid of Them and Save Memory
• String deduplication feature (from Java 8 update 20)
• G1 GC: Reducing Memory Consumption by 20%