[OpenJDK 2D-Dev] New GeoServer benchmarks with Laurent's l4 webrev

[Andrea Aime]

Hi all,
upon Laurent request I've re-run the benchmarks including more data points,
that is,
adding the ConcurrentLinkedQueue (CLQ) caching as an alternative to the
Thread Local
(TL) caching, and running all benchmarks again with an output format which
is not
as taxing as PNG, in particular JPEG.
The latter moves us out of a real world test, because the maps in question
are mostly
vector, nobody would compress them in JPEG due to the compression artifacts
being
quite visible in large areas of solid color, but we can use to to "imagine"
what the
differences would be if we had a faster PNG encoder.

Here are the results for the PNG case, same as before, but with the extra
CLQ column:

  *Threads* *Vanilla* *L1 – TL* *L4 – TL* *L4 – CLQ*  1 12,20 13,12 13,09
13,34  2 20,16 21,04 21,45 21,66  4 32,29 34,88 35,08 33,98  8 43,30 44,71
44,00 45,05  16 47,36 49,49 50,99 51,04  32 44,19 50,55 52,11 49,83  64
42,77 44,56 47,31 47,80
As you can see CLQ is pretty close, the differences we see are imho mostly
due to
the natural variation between tests. Given that overall CLQ has a better
memory
usage profile I'd say it is rather interesting at least for server side
usage.

And here are the results for a JPEG benchmark, where the cost of the
encoding is significantly lower:

  *Threads* *Vanilla* *L1 – TL* *L4 – TL* *L4 – CLQ*  1 20,42 22,63 23,56
23,55  2 33,56 34,81 35,35 36,00  4 49,68 56,83 57,39 55,90  8 62,58 65,27
67,96 70,20  16 70,67 74,92 76,88 81,05  32 65,30 74,56 74,61 75,19  64
60,59 63,73 69,26 69,57


This benchmarks confirms the performance goodness of Laurent's
work (if there still was any need for that...) and shows, interestingly,
an advantage for the CLQ solution under high load.

Pretty good stuff... go Laurent!

Now... if only I can manage to write a faster PNG encoder :-p

Cheers
Andrea

-- 
==
Our support, Your Success! Visit http://opensdi.geo-solutions.it for more
information.
==

Ing. Andrea Aime
@geowolf
Technical Lead

GeoSolutions S.A.S.
Via Poggio alle Viti 1187
55054  Massarosa (LU)
Italy
phone: +39 0584 962313
fax: +39 0584 1660272
mob: +39  339 8844549

http://www.geo-solutions.it
http://twitter.com/geosolutions_it

-------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/2d-dev/attachments/20130622/9c43bf14/attachment.html>