Why is the native String getBytes method slower than the custom implemented getBytesFast?

Question

When running the following piece of code, the execution of the Java String's native method getBytes() seems to be slower than the custom getBytesFast() implementation. You can use the Arrays.equals(str.getBytes(), getBytesFast(str)) to verify that both byte arrays are equals.

The getBytesFast implementation is a modified version of the implementation included in this programming tips article (1997): http://java.sun.com/developer/technicalArticles/Programming/Performance/

I'm looking for a well documented answer on why the native implementation is slower than the custom implementation.

package com.test;

public class Performance {

    public static void main(String args[]) {

        final String str = "This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test! This is a performance test!";

        long startTime_1 = System.nanoTime();
        str.getBytes();
        System.out.println(System.nanoTime() - startTime_1);

        long startTime_2 = System.nanoTime();
        getBytesFast(str);
        System.out.println(System.nanoTime() - startTime_2);
    }

    private static byte[] getBytesFast(String str) {
        final char buffer[] = new char[str.length()];
        final int length = str.length();
        str.getChars(0, length, buffer, 0);
        final byte b[] = new byte[length];
        for (int j = 0; j < length; j++)
            b[j] = (byte) buffer[j];
        return b;
    }
}

EDITED:

Caliper benchmark results

Thanks

Answer 1

According to [the documentation], the char type in Java is a 16-bit Unicode character, whereas the byte type is an 8-bit signed integer. This means that with each char-to-byte cast made in your code you're throwing away half the character data.

The Java tutorial on Character and Byte Streams has a nice little example string using Japanese Kanji:

String jaString = new String("\u65e5\u672c\u8a9e\u6587\u5b57\u5217");

For each character in that string your fast conversion method would throw away the first byte of information (e.g. the 65 in \u65e5). Your link also specifically mentions that String.getBytes() runs several times more slowly "because the former does correct byte-to-char conversion, which involves a function call per character."

If you completely ignore character encoding and throw away the higher-order byte from each char then you'll get a bit of a speedup. You just need to keep in mind that this method only works with certain character encodings and has a potential for data loss.

Answer 2

String.getBytes takes into account the default charset of the system. Your implementation assumes ISO-8859-1.

String.getBytes eventually ends up calling this method. ce is a CharsetEncoder.

byte[] encode(char[] ca, int off, int len) {
        int en = scale(len, ce.maxBytesPerChar());
        byte[] ba = new byte[en];
        if (len == 0)
            return ba;
        if (ce instanceof ArrayEncoder) {
            int blen = ((ArrayEncoder)ce).encode(ca, off, len, ba);
            return safeTrim(ba, blen, cs, isTrusted);
        } else {
            ce.reset();
            ByteBuffer bb = ByteBuffer.wrap(ba);
            CharBuffer cb = CharBuffer.wrap(ca, off, len);
            try {
                CoderResult cr = ce.encode(cb, bb, true);
                if (!cr.isUnderflow())
                    cr.throwException();
                cr = ce.flush(bb);
                if (!cr.isUnderflow())
                    cr.throwException();
            } catch (CharacterCodingException x) {
                // Substitution is always enabled,
                // so this shouldn't happen
                throw new Error(x);
            }
            return safeTrim(ba, bb.position(), cs, isTrusted);
        }
    }
}

private static int scale(int len, float expansionFactor) {
    // We need to perform double, not float, arithmetic; otherwise
    // we lose low order bits when len is larger than 2**24.
    return (int)(len * (double)expansionFactor);
}

private static char[] safeTrim(char[] ca, int len,
                               Charset cs, boolean isTrusted) {
    if (len == ca.length && (isTrusted || System.getSecurityManager() == null))
        return ca;
    else
        return Arrays.copyOf(ca, len);
}

There is a much larger degree of complexity involved with using a CharsetEncoder, which could account for the slower execution times you are seeing.

Answer 3

It may be because String.getBytes() uses or delegates to a Charset (the JVM's current default one) and your "fast" implementation is just hard-coded ISO-8859-1 charset.

(Note: i did not verify your results, i'm just stating my hipothesis here. Comments relating micro benchmarks are more than important here, and definitely more valuable than my answer to your question :)