Re: [Vtd-xml-users] Performance in comparison to libxml2

[Jimmy Z. <cra...@co...>]

Tatu, we have done some internal testing, Java and C version are very close 
in
performance for most cases, when XML is small, C out performs Java version.
The other thing: by java what I mean is using server JVM (hotspot) which 
does a
phenominal job doing JIT compiler... An other observation: C# is the 
slowest,
20% slower than Java and C...
Anyway, we are doing a version 1.8 release, which should come out pretty 
soon,
the big feature is XMLModifier, which is a easy to use API that allows one 
to take
advantage of VTD's incremental update capability, by then, we will try to 
post
updated benchmark results includes side-by-side comparison of C and Java...
Cheers,
Jimmy

----- Original Message ----- 
From: "Tatu Saloranta" <cow...@ya...>
To: "Jimmy Zhang" <cra...@co...>; "John Kraal" <jk...@in...>
Cc: <vtd...@li...>
Sent: Tuesday, November 07, 2006 11:29 AM
Subject: Re: [Vtd-xml-users] Performance in comparison to libxml2


> --- Jimmy Zhang <cra...@co...> wrote:
>
>> The performnace test doesn't look right to me...
> ...
>> I parsed a 170MB today on windows (1.7 ghz pentium
>> M)
>> running XML parsing  it took about 3 seconds to
>> parse that
>
> Hi Jimmy! Besides comparing libxml2 and vtd-xml/c, it
> would be quite interesting to see how vtd-xml c and
> java implementations differ regarding performance. I
> know how to measure Java version, but since it's been
> a while I did anything in c/c++, I haven't had time to
> figure out if performance differs significantly
> (probably not, but who knows?)
>
> So... would it be possible for you to maybe run some
> side-by-side comparisons between Java and C
> implementations of vtd-xml? These would be quite
> enlightening. Also, for c version, it'd be good to do
> similar warmup as is done for Java version, to try to
> minimize startup costs (which are much less of a
> concern for native code... but there's still some
> overhead for the first run).
>
> I suspect original poster should also try running the
> same/similar parsing multiple times per test -- just
> running executable once may give skewed results (maybe
> there are lib deps to load or such). Especially when
> runtime is very short, couple of seconds.
>
> -+ Tatu +-
>
>
>
>
>
>
> ____________________________________________________________________________________
> Want to start your own business?
> Learn how on Yahoo! Small Business.
> http://smallbusiness.yahoo.com/r-index
>