Rhino over Duktape

[HugoGresse]

Hi,

I was searching for a JS engine for Android and it seems that Rhino is ~10% more efficient than Duktape. What do you think about it? Maybe a real benchmark could be helpful...

[swankjesse]

Please do! Benchmarking on Android would be particularly useful since Rhino relies on the host VM whereas Duktape doesn’t.

I doubt we’d switch regardless of the numbers because we’re optimizing for compactness and not throughput.

[HugoGresse]

for what I see, runing 1000x the same operation which is creating a js function that return 1 and calling it and counting the return value in java, so the whole java > js > java chain is 10% faster using Rhino. What's more, Rhino is 600ko once packaged, Duktape is 2mB. To finish, Rhino is java based, no JNI which reduce the crash potential drastically.

DUKTAPE

    public static final String COMMAND ="function runTest() { return 1; }\n" +
            "var result = runTest();\n";
    Duktape mDuktape;

    public DuktapeTest(Listener listener, int loopNumber, Context context) {
        super(listener, loopNumber, context);

        mDuktape = Duktape.create();

        mDuktape.evaluate(COMMAND);


        long before = System.nanoTime();
        int result = 0;

        try {
            for(int i = 0; i < mLoopNumber; i++){
                result += test();
            }
        } finally {
            mDuktape.close();
        }

        if(result != mLoopNumber) {
            Log.w("Duktape", "Error loop end not reached correctly");
            super.onResult(-1);
            return;
        }

        super.onResult(System.nanoTime() - before);
    }

    private double test(){
        return Double.parseDouble(String.valueOf(mDuktape.evaluate("runTest()")));
    }

RHINO

    org.mozilla.javascript.Context mJSContext;

    public RhinoJarTest(Listener listener, int loopNumber, Context context) {
        super(listener, loopNumber, context);

        mJSContext = org.mozilla.javascript.Context.enter();

        mJSContext.setOptimizationLevel(-1);

        Scriptable scope = mJSContext.initStandardObjects();

         mJSContext.evaluateString(scope, "function runTest() { return 1; }", "ScriptAPI", 1, null);

        long before = System.nanoTime();
        int result = 0;

        try {
            for(int i = 0; i < mLoopNumber; i++){
                result += test(scope);
            }
        } finally {
            org.mozilla.javascript.Context.exit();
        }

        if(result != mLoopNumber) {
            Log.w("Rhino", "Error loop end not reached correctly");
            super.onResult(-1);
            return;
        }

        super.onResult(System.nanoTime() - before);
    }

    private double test(Scriptable scope){
        return Double.parseDouble(String.valueOf(mJSContext.evaluateString(scope, "runTest()", "ScriptAPI", 1, null)));
    }

[swankjesse]

Neat! Which OS / device? Mind providing the specific numbers? Also what's the time elapsed to run a single operation?

[swankjesse]

(I’m concerned about the potential time to initialize the JS engine. The 10% faster thing means less if it's 20% slower to initialize!)

[HugoGresse]

I'll probably publish a repo with the benchmarks (how to execute JS in Android). Anyways, here is a result. I should probably run some real average on the result and complete the JS code with real calculation (not just returning values). Anyways, the AndroidJSCore is pretty bad.

[HugoGresse]

It didn't count the initialize time, but based on my eye duktape and Rhino initialize time are pretty correct compared to webview for example.

Moto X 2014 android 5.1

[swankjesse]

Cool. Looking forward to seeing your repo with benchmarks.