Aspose 11 has performance downgrade a lot compared to Aspose 10.5

Aspose.Words Product Family
21

Hey Konstantin,

>>> Is it
performance of 11.1.1 is good enough?
No it is not. My test results should make that clear.

>>> My 11.1.1 performs better than 10.5 in any of my environments.
If I’m interpreting your results correctly the 3rd set of tests (jdk 1.5 w/o sections) show a performance degration between 11.1 and 10.5

The major difference is the JDK used. We have IBM JDK while your tests were made with Oracle’s JDK. The absolute values are not interesting as the hardware differs but the difference between Aspose 10.05 and Aspose 11.1 times.

Regards,
Dragos

22

Quick update: introducing a section break every 500-1000 paragraphs results in a huge performance improvement for my 15000 elements tests, 13-20 seconds down from 366 seconds.

Regards,
Dragos

23

Hey,

another update: using section breaks ( any type of section break but not page breaks) helps with performance even with Aspose 11.0. My 15000 element test executes in 12 seconds now if a section break is introduced every 200 pages compared to the 366 seconds needed without a section break.

With section breaks the results are identical with the ones obtained for Aspose 11.1.11. So the original issue reported can be narrowed down to the single section scenario and that 11.1.11 brings no improvements for it.

Regards,
Dragos

24

Hi Dragos,

Just a quick point but in Konstantin’s tests his 11.1.1 is typically faster than 10.5 so obviously from our point of view our latest build outperforms 10.5 which is leading to the confusion.

To my knowledge all our testing is based on Oracle JDK as the most widely used among our customers, while we have had a few issues raised before specific to IBM JDK I do not think they’ve been that many to justify running separate tests. Let me confirm this internally and I’ll come back to you.

Best Regards,

Billy.

25

Hi Dragos,

My 3rd row shows test of AW 10.5 under jdk1.5 and w/o sections. It can be compared to 6th row – AW11.1.1 under jdk1.5 and w/o sections. Comparison of this two rows shows that original test (w/o sections) under 11.1.1 is about 12% faster than 10.5 (in my environment).

IBM JIT compiler is awesome, but sometimes it introduces surprises, not always good ones.

AW optimized for small or middle number of paragraphs in a section – as in most real-world documents. Documents with very big sections (1000 paragraphs and more) are processed considerably slower in any AW version.

And I didn’t understand: is performance of 11.1.1 with section breaks compatible with performance of 10.5 on your environment?

Regards,

26

@Billy - I have not performed tests with the Oracle JVMso I cannot provide results there but Konstantin’s results ( or at least how I read them) seem to indicate a performance decrease for the Oracle JVM as well.

@Konstantin,

>>>My 3rd
row shows test of AW 10.5 under jdk1.5 and w/o sections. It can be compared to
6th row – AW11.1.1 under jdk1.5 and w/o sections. Comparison of this
two rows shows that original test (w/o sections) under 11.1.1 is about 12%
faster than 10.5 (in my environment).
I am looking at these results from your table which show a decrease for 11.1. Am I reading the table wrong? See column 3 and column 6, column 6 has higher values.

AW 10.05, JDK 1.5, w/o sections: 9 20 38 77 139
AW 11.1, JDK 1.5, w/o sections: 7 18 44 127 1186

>>> IBM JIT
compiler is awesome, but sometimes it introduces surprises, not always good
ones.
That might be the case but I’m reporting the results to 10.05 where, with the exact same compiler and test environment the results are way better.

>>> AW
optimized for small or middle number of paragraphs in a section – as in most
real-world documents. Documents with very big sections (1000 paragraphs and
more) are processed considerably slower in any AW version.
It looks like this indeed. But my question is why AW 10.05 was doing a good job even for this scenario? What has changed in AW 11.1?

>>> And I didn’t
understand: is performance of 11.1.1 with section breaks compatible with
performance of 10.5 on your environment?

Running 11.1.11 with section breaks every 200 paragraphs yeilds similar results to AW 10.05 w/o section breaks. Increasing the paragraphs/section decreases performance in AW 11.1.11. I assume that if I would run AW 10.05 with section breaks I might see better results yet but the difference would not be as high as for the w/o section breaks scenario.

Regards,
Dragos

27

Hi Dragos,

Please check just released 11.2 – it contains ultimateJ fix and shows linear performance for documents without section breaks. You can get it from: https://downloads.aspose.com/words/java.

About interpretation of table. We look at it from different ends: I – from begin, you – from end:). Time of the test is hardly depended by a heap size. This is “gray” area of heap size just before heap error: application still works (doesn’t throw heap error) but performance is non-linear because of “swapping” effect. If you raise available heap size for the application this gray area is moved further. So I counted only the first half of table when spoke about improved performance.

Best regards,

28

The issues you have found earlier (filed as WORDSJAVA-510) have been fixed in this .NET update and this Java update.