Ok people,
I just tried a quick test on my laptop with the patched XSLTC that Jacek
sent me. The results are outstanding!
I ran just one test, the rest isn't useful anymore:
XSLTC (patched) on JDK 1.3.1 yields 590,44
!!!!!!!!!!!
In case you don't remember, MSXML 3.0 (natively) yields more or less 320
So, in case math is not your friend, this says that XSLTC is *twice* as
fast as a native processor on the same machine!!!!
So, Xalan, what about giving Jacek CVS commit access so that he can
continue to work on XSLTC and improve it's performance? :)
Ah, BTW, let me file my first bug report for XSTLC: the XSLTMark
'number' test gives me the following stacktrace:
number: number.xsl number.xml ...
Attempting to format number '22.0' using pattern ##,##,00.##'.
java.lang.RuntimeException: Attempting to format number '22.0' using
pattern '##,##,00.##'.
at
org.apache.xalan.xsltc.runtime.BasisLibrary.runTimeError(BasisLibrary.java:999)
at
org.apache.xalan.xsltc.runtime.BasisLibrary.runTimeError(BasisLibrary.java:1007)
at
org.apache.xalan.xsltc.runtime.BasisLibrary.formatNumber(BasisLibrary.java:884)
at xsltmark24.number.applyTemplates()
at xsltmark24.number.applyTemplates()
at xsltmark24.number.applyTemplates()
at xsltmark24.number.transform()
at
org.apache.xalan.xsltc.runtime.AbstractTranslet.transform(AbstractTranslet.java:547)
at
com.datapower.xsltmark.XSLTCDriver.runTransform(XSLTCDriver.java:200)
at
com.datapower.xsltmark.XSLTMark.runBenchmark(XSLTMark.java:96)
at com.datapower.xsltmark.XSLTMark.main(XSLTMark.java:506)
transform failed!
Also, XSLTMark reports these results
Results for driver XSLTCDriver
prep exec
Test Component iter ms ms KB In KB Out KB/s
status
alphabetize 10 210 661 196 0 149,07 CHK
OUTPUT
attsets 100 140 732 121 256 257,75 OK
avts 100 90 1462 1966 1157 1068,07 OK
axis 100 200 501 35 61 96 OK
backwards 100 90 1963 260 261 133,05 OK
bottles 100 311 1312 4 1130 432,49 OK
breadth 100 80 821 415 200 375,64 OK
brutal 100 231 1131 130 536 295 OK
chart 100 111 1291 121 604 281,13 OK
creation 50 100 2143 983 1520 584,21 OK
current 100 90 411 18 30 59,52 OK
dbonerow 10 100 1802 19621 11 5447,44 OK
dbtail 100 80 941 1966 504 1312,44 OK
decoy 100 150 5909 1966 1974 333,45 OK
depth 100 40 891 588 373 540,23 OK
encrypt 10 70 681 196 196 288,81 OK
functions 100 110 6159 1966 618 209,82 OK
game 100 110 481 260 71 344,94 OK
html 100 100 411 29 77 129,5 CHK
OUTPUT
identity 10 60 1773 1962 1962 1106,93 OK
inventory 100 161 691 184 202 280,04 OK
metric 100 170 881 88 65 87,46 CHK
OUTPUT
number 0 110 0 0 0 ?
ERROR
oddtemplate 100 80 300 16 15 54,04 OK
patterns 100 101 3164 1966 1974 622,74 OK
prettyprint 10 131 6779 196 1624 134,3 OK
priority 100 130 350 91 57 212,05 OK
products 100 80 761 196 55 165,09 OK
queens 10 200 1342 0 17 6,53 OK
reverser 100 100 631 129 129 205,53 OK
stringsort 10 71 2804 1962 1933 694,7 OK
summarize 100 111 741 712 29 500,67 CHK
OUTPUT
total 100 90 290 121 45 287,75 OK
tower 20 211 4226 1 1381 163,57 OK
trend 100 100 9714 176 806 50,61 CHK
OUTPUT
union 100 90 271 16 11 53,15 OK
xpath 100 70 270 28 20 91,15 OK
xslbench1 100 130 1022 36 809 413,84 OK
xslbench2 20 80 1903 2920 2290 1369,26 OK
xslbench3 20 60 210 2920 25 7014,32 OK
Aggr. Results 2570 3948 55418 43374 22066 590,44
Arithmetic Mean 116,12 733,38
Geometric Mean 300,86
Harmonic Mean 107,97
Conformance Score (OK+NOREF+SKIPPED divided by TOTAL) 34/40 = 85%
which means that
- alphabetize
- html
- summarize
- trend
deviate from what Xalan and Saxon do (so might be a conformance error).
Question: do you Xalan guys run the conformance tests on XSLTC as well?
Now, Jacek also patched XSTLC to make it work with Cocoon.... I'm on to
try that out right now. I'll report back when I have more info.
For now, I'm *very* happy to see *finally* java shining performance-wise
:)
I think this shows that using a native XSLT processor for Cocoon isn't
such a great idea after all. Happy to find it out before attempting to
make one work :)
And remember: These are results using hotspot client technology....
hotspot server further compiles things down as you go but you need much
more runs to appreciate them... better tests will come out for real-life
use on the server.
--
Stefano Mazzocchi One must still have chaos in oneself to be
able to give birth to a dancing star.
<[EMAIL PROTECTED]> Friedrich Nietzsche
--------------------------------------------------------------------
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, email: [EMAIL PROTECTED]
