[OpenSTA-devel] Re: Duration of the viewstate encoding

[Daniel S. <da...@Op...>]

Thierry Boullet wrote:
> I made some measurements of the duration of the viewstate
> encoding because I noted that it is not negligible and that it can
> disturb the tests.

There's no argument that doing SCL string encoding is inneficient and
that we need to add a way of doing this efficiently in SCL...

However this mail isn't about a native  SCL encoding method, instead
it's about the benchmarks that Thierry created and the possibility of
speeding OpenSTA SCL string handling up by building it differently.

I have taken Thierry's URL encoding methods, and his viewstate data,
added a third encoding method (a really inneficient one) and packaged
them up so they can be run by anyone, anywhere.

The SCL mechanism Thierry referred to as METHOD2 (the method Thierry
uses) has become my URLENCODE1
The SCL mechanism Thierry referred to as METHOD1 (the FAQ method) has
become my subroutine URLENCODE2
The SCL mechanism I created (a simple one which is written without
efficiency in mind) has become my subroutine URLENCODE3

The script that contains these 3 subroutines reads the viewstate data
from a .FVR data file and times each subroutine for each of the
viewstates, it doesn't do any HTTP.  You can download this test here:
  http://ftp.opensta.org/tests/urlencode-050620.zip
I ran the script in a test so it iterated 20 times on a machine that
had been freshly rebooted each time.

The test machine was a P4/2.8GHz with 1Gb RAM running WinXPproSP2.  I
ran the tests first with the stock 1.4.3 release and then again with
a CVS source build, built against the WinServer2003SP1 PSDK and
STLport 4.6.2 - the other externals are as recommended for 1.4.3, the
build needed some small "adjustments" to OmniORB and the STLport to
work cleanly...  STLport was set up to use its own IOSTREAMS.

The results (just like Thierry's) are pretty conclusive:
                      1.4.4        New
                     Release      Build
      /  URLENCODE1    5.51        1.82
VSTATE1  URLENCODE2    7.21        2.17
      \  URLENCODE3  180          70.1

      /  URLENCODE1    1.26        0.40
VSTATE2  URLENCODE2    1.38        0.42
      \  URLENCODE3  189          68.2

I'm working on some other builds at the moment - one that is similar
to the above build but where the STLport is set up to just provide
shells around the platform iostreams implementation; another doesn't
use the STLport at all.  I'd like to try an STLport 5.0 build but
it's driving me insane ... ;-)

Thoughts, comments, etc. all welcome
Cheers
/dan
-- 
Daniel Sutcliffe <Da...@Op...>
  OpenSTA part-time caretaker - http://OpenSTA.org/