Performance: Nashorn vs. Node

I am naively playing around with Nashorn. Nashorn is the new JavaScript Runtime that comes with Java 8 and runs on the JVM.

I am using the current Build b132 of the Early Access Release of Java 8 on a recent MacBook Pro (2.3 GHz Intel Core i7)

One of my first experiments is comparing performance with node.js.

I have run the following script with Nashorn and with Node:

The results:

Nashorn: ✔ jjs fib.js Input: 30, Time: 66 Input: 31, Time: 40 Input: 32, Time: 58 Input: 33, Time: 90 Input: 34, Time: 154 Input: 35, Time: 242 Input: 36, Time: 348 Input: 37, Time: 556 Input: 38, Time: 920 Input: 39, Time: 1483 Input: 40, Time: 2426 Input: 41, Time: 3928 Input: 42, Time: 6360 Input: 43, Time: 10303 Input: 44, Time: 16602 Input: 45, Time: 26802 Input: 46, Time: 43384 Input: 47, Time: 72201 Input: 48, Time: 116425 Input: 49, Time: 188520

Node: ✔ node fib.js Input: 30, Time: 13 Input: 31, Time: 20 Input: 32, Time: 33 Input: 33, Time: 53 Input: 34, Time: 88 Input: 35, Time: 142 Input: 36, Time: 226 Input: 37, Time: 365 Input: 38, Time: 590 Input: 39, Time: 954 Input: 40, Time: 1530 Input: 41, Time: 2522 Input: 42, Time: 4047 Input: 43, Time: 6534 Input: 44, Time: 10557 Input: 45, Time: 17161 Input: 46, Time: 27712 Input: 47, Time: 46323 Input: 48, Time: 73521 Input: 49, Time: 118905

Pure Java: ✔ java -cp fibdemo.jar FibDemo Input: 30, Time: 5 Input: 31, Time: 6 Input: 32, Time: 11 Input: 33, Time: 18 Input: 34, Time: 27 Input: 35, Time: 45 Input: 36, Time: 71 Input: 37, Time: 121 Input: 38, Time: 188 Input: 39, Time: 304 Input: 40, Time: 490 Input: 41, Time: 804 Input: 42, Time: 1276 Input: 43, Time: 2067 Input: 44, Time: 3361 Input: 45, Time: 5445 Input: 46, Time: 8901 Input: 47, Time: 14253 Input: 48, Time: 23288 Input: 49, Time: 37160

Out of the box the performance of Node seems to be almost a factor 1.6 better than of Nashorn.

Of course this naive experiment is not representative for real world performance and this is still an early access release of Java 8. It might well be that the performance of the final release will be better. Also the JVM can probably be tuned in many ways unknown to me.

Still I would be interested in thoughts about the findings of this experiment...