Is it really faster than Codecutils??

[tmdgusya]

Describe the bug

https://github.com/aws/aws-sdk-java/blob/8542b3ddc6be843dc910b3ccb1e6aeb3cc05c001/aws-java-sdk-core/src/main/java/com/amazonaws/util/Base64.java

In Base64 files, you described that if you don't use JAXB, then will fallback to SDK implementation which may be less performant. So, I tested this. I tested which is the faster one between CodecUtils and DataTypeConverter. As described comment on function, DataTypeConverter should be faster than CodecUtils. But, DataTypeConverter always slower than CodecUtils.

Expected Behavior

DataTypeCoverter is better than CodecUtils. (But Always lose)

Current Behavior

DataTypeCoverter is always slower than CodecUtils.

Reproduction Steps

class LearningTest : FunSpec({
    test("which is faster?") {
        val codec = Base64Codec()
        // initialize Base64 Class because we have to execute static block code before testing
        Class.forName("com.amazonaws.util.Base64")
        Class.forName("javax.xml.bind.DatatypeConverter");


        // 1. DatatypeConverter
        val usingDatatypeConverter = measureNanoTime {
            DatatypeConverter.printBase64Binary(byteArrayOf(
                -86,
                24,
                85,
                104,
                67,
                51,
                80,
                -80,
                -68,
                -48,
                -35,
                28,
                1,
                -92,
                -43,
                110
            ))
        }

        val hello = measureNanoTime { println("Hello") }

        // 2. CodecUtils.toStringDirect
        val usingCodeUtils = measureNanoTime {
            CodecUtils.toStringDirect(
                codec.encode(
                    byteArrayOf(
                        -86,
                        24,
                        85,
                        104,
                        67,
                        51,
                        80,
                        -80,
                        -68,
                        -48,
                        -35,
                        28,
                        1,
                        -92,
                        -43,
                        110
                    )
                )
            )
        }


        println("usingDatatypeConverter: $usingDatatypeConverter") // 6455000
        println("usingCodeUtils: $usingCodeUtils") // 386000
        println("hello: $hello") // 30417

        // usingBase64 is faster than usingCodeUtils (fail)
        usingCodeUtils.shouldBeGreaterThan(usingDatatypeConverter)
    }
})

Possible Solution

No response

Additional Information/Context

    implementation("com.amazonaws:aws-java-sdk-s3:1.12.174")
    implementation("javax.xml.bind:jaxb-api:2.3.1")
    implementation("com.amazonaws:aws-java-sdk-s3")

AWS Java SDK version used

1.12.174

JDK version used

17

Operating System and version

macOS monterey 12.6 (21G115) M1Max Memory: 32GB

[debora-ito]

@tmdgusya what is the ask here?

[tmdgusya]

@debora-ito

            LOG.warn("JAXB is unavailable. Will fallback to SDK implementation which may be less performant." +
                     "If you are using Java 9+, you will need to include javax.xml.bind:jaxb-api as a dependency.");

    public static String encodeAsString(byte... bytes) {
        if (bytes == null) {
            return null;
        }
        if (isJaxbAvailable) {
            try {
                return DatatypeConverter.printBase64Binary(bytes);
            } catch (NullPointerException ex) {
                // https://netbeans.org/bugzilla/show_bug.cgi?id=224923
                // https://issues.apache.org/jira/browse/CAMEL-4893

                // Note the converter should eventually be initialized and printBase64Binary should start working again
                LOG.debug("Recovering from JAXB bug: https://netbeans.org/bugzilla/show_bug.cgi?id=224923", ex);
            }
        }

        return bytes.length == 0 ? "" : CodecUtils.toStringDirect(codec.encode(bytes));
    }

High performant(DataTypeConverter)
Less performant(CodecUtils)

According to the comment in this file, DataTypeConverter should be faster than CodecUtils.
However, when I tested it, CodecUtils was always faster than DataTypeConverter.

[YanivKunda]

@tmdgusya did you check that in a multi-threaded environment?