cxf-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benson Margulies" <bimargul...@gmail.com>
Subject Re: This caching/startup speedup project
Date Wed, 12 Mar 2008 21:27:33 GMT
Dan,

I also measured that case. The predominant cost is inside of XmlSchema. What
shall I do? Open a hailing frequency to the XmlSchema empire to discuss a
more integrated joint development model, so we can do things that depend on
work in their code in a predictable way?

I did find a fat 5% sitting in redundant classloader calls in the JaxWs
code, and I'm tuning a caching solution to that bit of the problem.

However, what would really make a difference to the average user is the
spring / xerces situation.

--benson



On Wed, Mar 12, 2008 at 4:27 PM, Daniel Kulp <dkulp@apache.org> wrote:

> On Wednesday 12 March 2008, Benson Margulies wrote:
> > The Christopher Chen email from February indeed is talking about
> > building a service from WSDL on the client side. I'll take a look, but
> > I have to say that this use-case has a Doctor-Doctor feeling to me,
> > with the pill being wsdl2java?
>
> That's really not the whole solution though.  Even if you run wsdl2java
> and have it generate code, by default, it still will grab the wsdl and
> use it.  That's kind of per-spec as things like the soap:address are not
> burned into the code.
>
> Yes, you CAN get it to not use the wsdl, but by default, it still will.
>
> Dan
>
>
> > On Wed, Mar 12, 2008 at 2:57 AM, Benson Margulies
> > <bimargulies@gmail.com>
> >
> > wrote:
> > > So, I'm reading Dan D's blog:
> > >
> > >
> > > http://netzooid.com/blog/2007/07/03/tips-for-improving-springs-start
> > >-up-performance/
> > >
> > > Did we do these things? We still load a pile of XML files, and we
> > > may still be running schema validation, and I'm not sure where we
> > > stand in bean-volume.
> > >
> > >
> > >
> > >
> > > On Fri, Mar 7, 2008 at 11:45 AM, Benson Margulies
> > > <bimargulies@gmail.com>
> > >
> > > wrote:
> > > > Permit me to be really confused. Do people really expect speedy
> > > > performance when digesting a WSDL at runtime, as opposed to having
> > > > generated annotated code and then compiling it?
> > > >
> > > > On Fri, Mar 7, 2008 at 10:10 AM, Daniel Kulp <dkulp@apache.org>
> wrote:
> > > > > Most likely, you should grab one of the simple demos, start the
> > > > > server,
> > > > > and profile the client.  Put the main method in a for loop to
> > > > > make the entire thing run 10 times or something so you can see
> > > > > where things are a
> > > > > bit more expensive.
> > > > >
> > > > > There are two things to consider:
> > > > > 1) The pure java first case: in this case, the RSFB just deals
> > > > > with the
> > > > > reflection stuff
> > > > >
> > > > > 2) The wsdl first cases: in this case the RSFB has to work with
> > > > > both the
> > > > > WSDL and the reflection stuff to get things matched up.  This
> > > > > may be quite a bit more expensive.  I don't really know.
> > > > >
> > > > > Dan
> > > > >
> > > > > On Thursday 06 March 2008, Benson Margulies wrote:
> > > > > > I took a first look with an eval copy of JProfiler while
> > > > > > waiting for response from them.
> > > > > >
> > > > > > All our startup is in the noise for a typical unit test. Do
I
> > > > > > need a really big service to see the problem that got us into
> > > > > > this discussion:?
> > > > > >
> > > > > >
> > > > > > On Wed, Mar 5, 2008 at 3:09 PM, Benson Margulies
> > > > > > <bimargulies@gmail.com>
> > > > > >
> > > > > > wrote:
> > > > > > > It's CPU that I'm proposing to hunt.
> > > > > > >
> > > > > > >
> > > > > > > On Wed, Mar 5, 2008 at 2:52 PM, Bharath Ganesh
> > > > > > > <bharathganesh@gmail.com>
> > > > > > >
> > > > > > > wrote:
> > > > > > > > I had done some memory profiling on CXF 2.0.2, quite
some
> > > > > > > > time back.  I think it was pretty stable except for
some
> > > > > > > > issues in PolicyEngine and HandlerChainInvoker which
are
> > > > > > > > fixed on the
> > > > >
> > > > > trunk.
> > > > >
> > > > > > > > As of now I can see only
> > > > > > > > this [1].
> > > > > > > > Please do let me know if you are suspecting a leak
at some
> > > > >
> > > > > place.
> > > > >
> > > > > > > > I shall
> > > > > > > > look into it.
> > > > > > > >
> > > > > > > > Thanks
> > > > > > > > Bharath
> > > > > > > >
> > > > > > > > [1]:
> > > > >
> > > > > http://www.nabble.com/Memory-leak-at-WSDLManagerImpl-td15131599.
> > > > >ht
> > > > >
> > > > > > > >ml
> > > > > > > >
> > > > > > > > On Wed, Mar 5, 2008 at 5:34 PM, Benson Margulies
> > > > > > > > <bimargulies@gmail.com>
> > > > > > > >
> > > > > > > > wrote:
> > > > > > > > > I might be willing to take a run at this. Can
someone
> > > > > > > > > tighten
> > > > >
> > > > > up
> > > > >
> > > > > > > > > the
> > > > > > > >
> > > > > > > > idea
> > > > > > > >
> > > > > > > > > a
> > > > > > > > > bit? If we were to cache service models, what
would we
> > > > > > > > > key/invalidate
> > > > > > > >
> > > > > > > > the
> > > > > > > >
> > > > > > > > > cache against? How do we know that we subsequently
need
> > > > >
> > > > > 'exactly
> > > > >
> > > > > > > > > the
> > > > > > > >
> > > > > > > > same
> > > > > > > >
> > > > > > > > > one'.
> > > > > > > > >
> > > > > > > > > For that matter, what about some profiling to
make sure
> > > > > > > > > that there
> > > > > > > >
> > > > > > > > isn't
> > > > > > > >
> > > > > > > > > something we could tighten up?
> > > > > > > > >
> > > > > > > > > Anyone have a favorite tool for that purpose?
> > > > >
> > > > > --
> > > > > J. Daniel Kulp
> > > > > Principal Engineer, IONA
> > > > > dkulp@apache.org
> > > > > http://www.dankulp.com/blog
>
>
>
> --
> J. Daniel Kulp
> Principal Engineer, IONA
> dkulp@apache.org
> http://www.dankulp.com/blog
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message