Asciidoctor.js on Nashorn JDK 9 is considerably slower than Nashorn JDK 8 (original) (raw)
Marcus Lagergren marcus.lagergren at oracle.com
Mon Mar 16 12:59:26 UTC 2015
- Previous message: RFR 8075207: Nashorn parser API returns StatementTree objects in out of order
- Next message: Asciidoctor.js on Nashorn JDK 9 is considerably slower than Nashorn JDK 8
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Yes. This probably has to do with optimistic typing being enabled to increase JavaScript runtime performance in 9. We know that 9 is currently slower at startup because we generate more correct code (and throw away too optimistic code that we generate first).
Try running with —optimistic-types=false and see if you get back to 8-like performance and startup.
We have plans to cheapen startup time considerably until 9 FC.
You can also try using the code caching feature from 8u40:
https://blogs.oracle.com/nashorn/entry/improving_nashorn_startup_time_using
Regards Marcus
On 03 Jan 2015, at 21:01, Guillaume Grossetie <grossetieg at gmail.com> wrote:
Hello everyone and happy new year! Asciidoctor.js is now running on both JDK 8b19 (build 25.40-b23) and JDK 9b44 (build 1.9.0-ea-b44). We noticed that JDK 9 is considerably slower than JDK 8 Nashorn but at the same time running "basic" Javascript seems faster on JDK 9. We have a simple Rake task[1] that execute the same examples with the latest JDK 8 and JDK 9: Result with JDK 8: Running Nashorn jjs spec/share/basic.js... OK in 909.052 ms Running Nashorn jjs spec/share/asciidoctor-convert.js... OK in 9335.534 ms Running Nashorn java BasicJavascriptWithNashorn... OK in 927.053 ms Running Nashorn java AsciidoctorConvertWithNashorn... OK in 8843.506 ms So less than a second for basic Javascript and around 10 seconds to convert AsciiDoc source to HTML5. Result with JDK 9: Running Nashorn jjs spec/share/basic.js... OK in 860.049 ms Running Nashorn jjs spec/share/asciidoctor-convert.js... OK in *31253.788 ms* Running Nashorn java BasicJavascriptWithNashorn... OK in 833.048 ms Running Nashorn java AsciidoctorConvertWithNashorn... OK in 30584.749 ms As you can see basic.js runs a little faster but asciidoctor-convert.js is considerably slower (three times slower ~30 seconds). The basic.js[2], asciidoctor-convert.js[3] and asciidoctor-all.js[4] are available on GitHub. Asciidoctor.js is generated from Ruby using Opal[5] as a result the file asciidoctor-all.js is not easy to read but the entry point is the method $convert[6]. Let me know if you need more information. Cheers, Guillaume [1] https://github.com/asciidoctor/asciidoctor.js/blob/master/Rakefile#L105 [2] https://github.com/asciidoctor/asciidoctor.js/blob/master/spec/share/basic.js [3] https://github.com/asciidoctor/asciidoctor.js/blob/master/spec/share/asciidoctor-convert.js [4] https://github.com/asciidoctor/asciidoctor.js/blob/master/dist/asciidoctor-all.js [5] http://opalrb.org/ [6] https://github.com/asciidoctor/asciidoctor.js/blob/master/dist/asciidoctor-all.js#L24419
- Previous message: RFR 8075207: Nashorn parser API returns StatementTree objects in out of order
- Next message: Asciidoctor.js on Nashorn JDK 9 is considerably slower than Nashorn JDK 8
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]