Re: [lmi] Calculation summary speed

[Evgeniy Tarassov]

On 10/23/06, Greg Chicares <address@hidden> wrote:
>           total  calculate prepare format
> old:        175        150       0     25 no xslt: all C++
> new:        950        150     300    500 current cvs
> modified:   430        150      30    250 cvs + patch below
>
> The "prepare" step became a decimal order of magnitude faster.
>
> Let's try to find a way to make the "format" step dramatically
> faster as well. That would be a big win. I'm not saying 250 ms
> is an excessive price to pay for flexibility, but I'd much
> rather pay less if at all possible.

After some profiling it appears that the 'format' step (on my
platform) breaks into:
320 milliseconds in total
----------
60 milliseconds in C++ code
260 milliseconds in xslt call

After disabling completly a double <xslt:for-each /> loop (the part
that generates the table), the xslt call time dropped to 17
milliseconds.

Currently the columns to be ouput are specified in the same template
html.xsl as a list of nodes and when the table is generated it adds
additional loop over that variable containing column names. If we i
remove that feature and explicitly specify the columns inside the
loop, then the time drops to 170 milliseconds.

But i think its not the final numbers and some performance improvments
could be achieved through our xsl templates optimisation. I will try
to google some additional information on how do they do optimize xslt
and will post back here.