Re: [Qexo-general] Speed of execution

[Eric Safern]

Per Bothner wrote:

> Eric Safern wrote:
>
> > With the full linux fortune data as its input (> 100,000 lines), the 
> > servlet takes 4 - 7 seconds on my test box, which strikes me as too long
> > for an impatient web surfer.  On a far smaller subset of data (4,000
> > lines), response time comes in at 1 - 2 seconds, which is still longer
> > than I'd like.
>
>
> Are you using a servlet?  My guess is that the 1-2 seconds is
> start-up time, and using a servlet can help with that, if you're
> not already using it.
>
> If you *are using a servlet, then the next step may be to pre-calculate
> document("/var/www/html/fortunes.xml")/fortunes/address@hidden/cookie
> for each $type, assuming there is only a small number of types.  You
> could either save this in separate files, or do this pre-calculation
> at server init type.  (The latter requires some Java programming,
> or at least some Scheme programming, since I don't think you can
> (currentlty do it in XQuery).  Or just pre-read the
> document("/var/www/html/fortunes.xml") in the server init
> code.
Well, before I spent too much time on coding, I ran a simple test to
see if the optimization would help - and it didn't seem to make much 
difference.

Here's what I did.  The current code loads a block of fortunes of a 
given type,
counts them, then generates a random number between 1 and (count), then
selects a specific fortune.

$cookies := 
document("/var/www/html/fortunes.xml")/fortunes/address@hidden/cookie,
       $min := 1,
       $max := count($cookies)

return  $cookies [ Rand($min,$max) ]

Per's suggestion is to pre-calculate the counts for the different types, 
so that
we can generate a random number between one and $count, and load only 
one fortune -
the specific one we wish to display.  So, as a test, I tried zeroing in 
on a specific fortune -
the fifth one (in my input data, no fortune type has less than 5 
fortunes, so this should be
safe for testing).


$cookies := 
document("/var/www/html/fortunes.xml")/fortunes/address@hidden/cookie[5],

return  $cookies

Note this test should run *faster* than the final code - I'm not loading 
the count data, or
generating the random number.

Unfortunately, it takes more or less the same time (six to eight 
seconds) to run the
servlet and display the results.  That is, of course, the time for the 
second and subsequent
runs - the first time is much longer, for obvious reasons.

So it doesn't seem like changing the algorithm will help much.  Am I 
missing something?

I'm not sure I understood your final suggestion.  If I pre-read the 
entire document in the server
init code, can I tell XQuery to run the query against the document *in 
memory*?  How
would that work?



Thanks,

Eric Safern