[graalvm-users] Benchmark

[Andreas Mueller]

Hi Francois,

unfortunately It seems that I might made a mistake… I don’t understand where, however...

This was an asynchronous test with request/reply (messaging). No network access, all inside the same JVM. To measure time I stamped every request with the current time and computed the delta when the reply came in. So I measured the latency. I don’t know why Nashorn had such a huge difference in latency over GraalVM.

I did some other tests today, async as well as sync. All request/reply. Unfortunately the results were quite similar. Both were able to handle about 60K requests per second. In some tests Nashorn beats GraalVM but not that much. I saw a bit more CPU usage in GraalVM and a longer warmup time for GraalVM until it reaches constant throughput.

I’m using JSR223 scripting engine. Except host access configured via Context for GraalVM there is no difference to the Nashorn engine. We use ES5. The JS are quite large and nested IIFEs (generated). The scripts run in different threads with engine scope. Each script within their own thread context. They communicate through queues that are provided from the Java part. All actions (events) are initiated from the Java part and then interface into the JS parts and from there back into Java and so fort.

I’d be glad to setup another test that might bring GraalVM in front of Nashorn.

Aside from this I’m quite happy with GraalVM. After a few changes I was able to run all our scripts with it. We will use GraalVM in our Docker distribution for the next release.

Regards,
Andreas
-- 
Andreas Mueller
IIT Software GmbH
https://urldefense.com/v3/__http://www.swiftmq.com__;!!GqivPVa7Brio!OTkOhMCgnkqvd7t-10OreduWGsgKGo6aeC4qf7UPdRTdVwZrbxidTID8-8RFRjHESTRQyg$  <https://urldefense.com/v3/__http://www.swiftmq.com__;!!GqivPVa7Brio!OTkOhMCgnkqvd7t-10OreduWGsgKGo6aeC4qf7UPdRTdVwZrbxidTID8-8RFRjHESTRQyg$ > 



On 25. Mar 2020, at 16:05, Francois Farquet <francois.farquet at oracle.com> wrote:

Hi Andreas,

This is a great news ! Going down from 25 minutes to less than a minute for the same workload, that's game changer.

If you happen to share your migration success story and/or your benchmark in a blog post or any other public way, please let us know !

Bests,

- Francois



On 25.03.20 15:50, Andreas Mueller wrote:
Hi,

I did a benchmark with a messaging application (quite a mix of JS/Java) on JDK 11.0.2/Nashorn and GraalVM 20.0.0. Multiple runs for warmup. Here are the best results:

On JDK 11.0.2/Nashorn: Duration 1’524’043 ms.

On GraalVM 20.0.0: Duration 47’813 ms.

Wow! It’s 31 times faster under GraalVM!

Thanks for providing this. It’s amazing!

Regards,
Andreas

-- 
Andreas Mueller
IIT Software GmbH
https://urldefense.com/v3/__http://www.swiftmq.com__;!!GqivPVa7Brio!OTkOhMCgnkqvd7t-10OreduWGsgKGo6aeC4qf7UPdRTdVwZrbxidTID8-8RFRjHESTRQyg$ 

<swiftmq_logo_positiv.png>



_______________________________________________
GraalVM-Users mailing list

GraalVM-Users at oss.oracle.com
https://oss.oracle.com/mailman/listinfo/graalvm-users
_______________________________________________
GraalVM-Users mailing list
GraalVM-Users at oss.oracle.com
https://oss.oracle.com/mailman/listinfo/graalvm-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://oss.oracle.com/pipermail/graalvm-users/attachments/20200326/d95c0c83/attachment.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: swiftmq_logo_positiv.png
Type: image/png
Size: 10241 bytes
Desc: not available
Url : http://oss.oracle.com/pipermail/graalvm-users/attachments/20200326/d95c0c83/attachment.png