[Java] Avoid unnecessary memory copy and addd a benchmark (#4611)

java/test/src/main/java/org/ray/api/benchmark/MicroBenchmarks.java

@@ -0,0 +1,48 @@
+package org.ray.api.benchmark;
+
+import org.ray.api.Ray;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+public class MicroBenchmarks {
+
+  private static final Logger LOGGER = LoggerFactory.getLogger(MicroBenchmarks.class);
+
+  public static Object simpleFunction() {
+    return null;
+  }
+
+  private static void time(Runnable runnable, int numRepeats, String name) {
+    LOGGER.info("Benchmark \"{}\" started.", name);
+    final long start = System.nanoTime();
+    for (int i = 0; i < numRepeats; i++) {
+      runnable.run();
+    }
+    final long duration = System.nanoTime() - start;
+    LOGGER.info(
+        "Benchmark \"{}\" finished, repeated {} times, total duration {} ms, average duration {} ns.",
+        name, numRepeats, duration / 1_000_000, duration / numRepeats);
+  }
+
+  /**
+   * Benchmark task submission.
+   *
+   * Note, this benchmark is supposed to measure the elapased time in Java worker, we should disable
+   * submitting tasks to raylet in `raylet_client.cc` before running this benchmark.
+   */
+  public static void benchmarkTaskSubmission() {
+    final int numRepeats = 1_000_000;
+    Ray.init();
+    try {
+      time(() -> {
+        Ray.call(MicroBenchmarks::simpleFunction);
+      }, numRepeats, "task submission");
+    } finally {
+      Ray.shutdown();
+    }
+  }
+
+  public static void main(String[] args) {
+    benchmarkTaskSubmission();
+  }
+}