Menu
▾
▴
Re: [Vtd-xml-users] Performance in comparison to libxml2
|
From: Jimmy Z. <cra...@co...> - 2006-12-05 08:14:55
|
will release the code and xml files into a zip file soon... ----- Original Message ----- From: "John Kraal" <jk...@in...> To: "Jimmy Zhang" <cra...@co...> Sent: Tuesday, December 05, 2006 12:02 AM Subject: Re: [Vtd-xml-users] Performance in comparison to libxml2 > I'll rebuild my programs and use your xml files for reference. > > Jimmy Zhang wrote: >> latest benchmarks >> >> http://vtd-xml.sf.net/benchmark.html >> >> >> ----- Original Message ----- >> *From:* John Kraal <mailto:jk...@in...> >> *To:* Jimmy Zhang <mailto:cra...@co...> >> *Sent:* Thursday, November 30, 2006 12:34 AM >> *Subject:* RE: [Vtd-xml-users] Performance in comparison to libxml2 >> >> Jimmy, >> >> I ran the tests multiple times to include startup and shutdown >> processing. I could create a loop doing 50 to 100 tests if that's >> what you prefer (or like to see). >> >> Creating strings is the actual retrieving of node data; looks quite >> important to me.. >> >> -----Original Message----- >> From: Jimmy Zhang [mailto:cra...@co...] >> Sent: Thu 30-11-2006 4:22 >> To: John Kraal >> Cc: vtd...@li... >> Subject: Re: [Vtd-xml-users] Performance in comparison to libxml2 >> >> Benchmarks usualy iterate the test for a large numbers before taking >> the >> average >> for each iteration, this is especially important for Java because >> server JVM >> takes >> some iterations to warm up ... >> Also Is there a need for creating strings?? A smart work around may >> exist to >> completely >> bypass this stage... >> ----- Original Message ----- >> From: "John Kraal" <jk...@in...> >> To: "Jimmy Zhang" <cra...@co...> >> Sent: Tuesday, November 28, 2006 11:48 PM >> Subject: Re: [Vtd-xml-users] Performance in comparison to libxml2 >> >> >> > In order >> > >> > * I used 1.8; but indeed I should've tested the java-cvs as well. >> > * Test code included; jkl.files.xml is described somewhat earlier >> (137 >> > mb) and the other attached as well. >> > * Java 1.5 something, it's a pain to get that installed and >> uninstalled >> > on Debian (Debian mindset: Proprietary software is evil) >> > * Parsing is actually reading the file and making it workable; >> maybe >> > that just didn't change significantly? >> > * I really cannot explain.. I believe it is now about the same >> speed as >> > '//*' >> > >> > Jimmy Zhang wrote: >> >> looks interesting.. a few things >> >> >> >> * you should check out the latest java version for the testing as >> well >> >> since there has been changes >> >> * What are the test code and conditions? We are doing a bit of >> >> benchmarking >> >> ourselves, mostly with modest file sizes... >> >> >> >> * for java version ,did you use server JVM, that makes a hell of >> >> difference?? >> >> >> >> I read it a bit more, there are more questions than answers... >> >> * from vtd-xml-c to vtdxml 1.8 cvs, there is no change at all >> >> at the parsing code, why there is a big difference in >> performance? >> >> >> >> * how come //*[local-name()="Descriptor"] takes shorter amoutn >> >> of time than //* >> >> ----- Original Message ----- From: "John Kraal" >> <jk...@in...> >> >> To: "Jimmy Zhang" <cra...@co...> >> >> Sent: Tuesday, November 28, 2006 4:38 AM >> >> Subject: Re: [Vtd-xml-users] Performance in comparison to libxml2 >> >> >> >> >> >>> I've tested the CVS version and included the results I've >> recorded until >> >>> now. I'm actually a bit too lazy to build other java-parsers to >> test >> >>> with, but it indicates a little of the C performance. >> >>> >> >>> I'm interested in your opinion on this before continuing with >> testing.. >> >>> >> >>> Regards, >> >>> John >> >>> >> >>> Jimmy Zhang wrote: >> >>>> There has been some significant XPath eval performance >> enhancement >> >>>> slated >> >>>> for next release, if you want to give a try, it is also in CVS >> moments >> >>>> ago... >> >>>> ----- Original Message ----- From: "John Kraal" >> <jk...@in...> >> >>>> To: "Jimmy Zhang" <cra...@co...> >> >>>> Sent: Tuesday, November 21, 2006 12:38 AM >> >>>> Subject: Re: [Vtd-xml-users] Performance in comparison to >> libxml2 >> >>>> >> >>>> >> >>>>> What failed exactly? Anyway, I'll do what I can; just notify >> me when >> >>>>> you >> >>>>> release 1.8. >> >>>>> >> >>>>> regards, >> >>>>> John >> >>>>> >> >>>>> Jimmy Zhang wrote: >> >>>>>> John, I wasn't able to get the toolchain script to work? >> >>>>>> Can you help us create another tar.gz release after the >> release of >> >>>>>> 1.8 >> >>>>>> in a few days.. >> >>>>>> Jimmy >> >>>>>> ----- Original Message ----- From: "John Kraal" >> <jk...@in...> >> >>>>>> To: "Jimmy Zhang" <cra...@co...> >> >>>>>> Cc: <vtd...@li...> >> >>>>>> Sent: Tuesday, November 07, 2006 12:12 AM >> >>>>>> Subject: Re: [Vtd-xml-users] Performance in comparison to >> libxml2 >> >>>>>> >> >>>>>> >> >>>>>>> No, I'm sorry - the files I used to test contains data from >> my >> >>>>>>> employer; >> >>>>>>> I will not be the one to distribute that :-). But it's quite >> easy to >> >>>>>>> create a fairly simple xml file like mine; and test it. >> >>>>>>> >> >>>>>>> Remember that I'm not testing a Java application, but merely >> C. >> >>>>>>> >> >>>>>>> The sources are in the previous email. >> >>>>>>> >> >>>>>>> The structure is quite simple, and as I'm not able to create >> a XSD >> >>>>>>> schema from a XML File easily (not really sure with what), >> here the >> >>>>>>> structure at it's simplest, just add some elements: >> >>>>>>> >> >>>>>>> <shipments> >> >>>>>>> <shipment dosvlg="123" afd="abc123"> >> >>>>>>> <from>NLAMS</from> >> >>>>>>> <to>NLRTM</to> >> >>>>>>> <goods> >> >>>>>>> <goodsline> >> >>>>>>> <merknr>abc</merknr> >> >>>>>>> <col code="pl">12345</col> >> >>>>>>> </goodsline> >> >>>>>>> </goods> >> >>>>>>> <relations> >> >>>>>>> <relation srtnaw="123" tsroln="123" relnr="123" zoek="abc"> >> >>>>>>> <tsnam1>abc</tsnam1> >> >>>>>>> <coordinates> >> >>>>>>> <tscoox>-12.3456</tscoox> >> >>>>>>> <tscooy>-12.3456</tscoox> >> >>>>>>> </coordinates> >> >>>>>>> </relation> >> >>>>>>> </relations> >> >>>>>>> </shipment> >> >>>>>>> </shipments> >> >>>>>>> >> >>>>>>> >> >>>>>>> >> >>>>>>> >> >>>>>>> Jimmy Zhang wrote: >> >>>>>>>> Can you provide a few sample XML files that you used for the >> >>>>>>>> testing? >> >>>>>>>> >> >>>>>>>> I am sure there can be additioal performance tuning for >> performance >> >>>>>>>> evaluation... >> >>>>>>>> >> >>>>>>>> So far the response we got compare VTD-XML favorably with >> Xerces... >> >>>>>>>> for the java version.. >> >>>>>>>> ----- Original Message ----- From: "John Kraal - Kewill >> >>>>>>>> Interchain NL" >> >>>>>>>> <jk...@in...> >> >>>>>>>> To: <vtd...@li...> >> >>>>>>>> Sent: Monday, November 06, 2006 4:23 AM >> >>>>>>>> Subject: [Vtd-xml-users] Performance in comparison to >> libxml2 >> >>>>>>>> >> >>>>>>>> >> >>>>>>>>> Hello, >> >>>>>>>>> >> >>>>>>>>> I've been playing with vtdxml for a while now, and my >> latest >> >>>>>>>>> experience >> >>>>>>>>> is not very encouraging to go on :(... >> >>>>>>>>> >> >>>>>>>>> I have created a file with _lots_ of recurring structures >> and >> >>>>>>>>> amounts of >> >>>>>>>>> recurring data (about 137 MB), eventually I created a xpath >> to >> >>>>>>>>> select >> >>>>>>>>> the name of every relation in the 4th role (doesn't really >> >>>>>>>>> matter). >> >>>>>>>>> >> >>>>>>>>> So, I executed it, with the code included for vtdxml (see >> >>>>>>>>> vtdxml.c), >> >>>>>>>>> and >> >>>>>>>>> it performed like this: >> >>>>>>>>> >> >>>>>>>>> unims@gxvm1:~/src/xmltest$ time ./test ./jkl.files.xml >> >>>>>>>>> '//relation[@tsroln=04]/tsnam1' >> /dev/null >> >>>>>>>>> >> >>>>>>>>> real 0m4.940s >> >>>>>>>>> user 0m4.625s >> >>>>>>>>> sys 0m0.256s >> >>>>>>>>> >> >>>>>>>>> With 162450 bytes of terminal output. >> >>>>>>>>> >> >>>>>>>>> Then, the horrible thing happened, I used lixml2 with an >> adjusted >> >>>>>>>>> reference xpath-program. (xpath1.c from their website, >> only with >> >>>>>>>>> content >> >>>>>>>>> retrieval): >> >>>>>>>>> >> >>>>>>>>> unims@gxvm1:~/src/libxmltest$ time ./test ./jkl.files.xml >> >>>>>>>>> '//relation[@tsroln=04]/tsnam1' >> /dev/null >> >>>>>>>>> >> >>>>>>>>> real 0m2.545s >> >>>>>>>>> user 0m2.201s >> >>>>>>>>> sys 0m0.321s >> >>>>>>>>> >> >>>>>>>>> The size of the output was 150808 bytes. >> >>>>>>>>> >> >>>>>>>>> This is an incredible difference, am I doing something >> wrong? Or >> >>>>>>>>> did I >> >>>>>>>>> have too many expectations of vtd? >> >>>>>>>>> >> >>>>>>>>> Regards, >> >>>>>>>>> John Kraal >> >>>>>>>>> >> >>>>>>>> >> >>>>>>>> >> >>>>>>>> >> -------------------------------------------------------------------------------- >> >>>>>>>>\ >> >>>>>>>>> >> ------------------------------------------------------------------------- >> >>>>>>>>> >> >>>>>>>>> >> >>>>>>>>> >> >>>>>>>>> Using Tomcat but need to do more? Need to support web >> services, >> >>>>>>>>> security? >> >>>>>>>>> Get stuff done quickly with pre-integrated technology to >> make your >> >>>>>>>>> job >> >>>>>>>>> easier >> >>>>>>>>> Download IBM WebSphere Application Server v.1.0.1 based on >> Apache >> >>>>>>>>> Geronimo >> >>>>>>>>> >> >> http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 >> >> <http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642> >> >>>>>>>>> >> >>>>>>>>> >> >>>>>>>>> >> >>>>>>>> >> >>>>>>>> >> >>>>>>>> >> -------------------------------------------------------------------------------- >> >>>>>>>> >> >>>>>>>> >> >>>>>>>> >> >>>>>>>> >> >>>>>>>> >> >>>>>>>> >> >>>>>>>>> _______________________________________________ >> >>>>>>>>> Vtd-xml-users mailing list >> >>>>>>>>> Vtd...@li... >> >>>>>>>>> https://lists.sourceforge.net/lists/listinfo/vtd-xml-users >> >>>>>>>>> >> >>>>>>>> >> >>>>>>>> >> >>>>>>> = >> > |
