Re: Tool that measures the performance of an XSLT prog

Thanks Michael. If there is no way to get the performance of the basic pieces
of an XSLT program, then how does one determine what is causing an XSLT
program to run slow? For example, I have a function f that calls functions A,
B, and C. The SAXON profile tool says that the total execution time for A, B,
and C is 1 ms. The profile tool says that total time for f is 6 minutes.
Without fine-grain profiling, I see no way to determine what is causing f to
take so long to execute. Suggestions?

/Roger

From: Michael Kay mike@xxxxxxxxxxxx <xsl-list-service@xxxxxxxxxxxxxxxxxxxxxx>
Sent: Sunday, July 26, 2020 4:19 PM
To: xsl-list@xxxxxxxxxxxxxxxxxxxxxx
Subject: [EXT] Re:  Tool that measures the performance of an XSLT program
ata fine granularity?

Short answer: no.

Even instrumenting the Java code internally is virtually impossible: the more
you try to probe timings at fine granularity, the more you distort things like
hot-spot optimization and CPU caching that have a profound effect on
performance. Heisenberg comes to mind...

Remember also that execution of these constructs isn't going to be sequential.
If the processor actually needs to do a run-time check that the results of
random:sequence are all xs:doubles, then it's probably going to check each
item as it's produced, rather than starting the checking when all the items in
the sequence are available. There's also lazy evaluation to consider: there's
a good chance that the processor will be evaluating the variable $hiddennodes
as a "side-effect" of one of the two expressions that references it.

Michael Kay
Saxonica


On 26 Jul 2020, at 20:36, Dr. Roger L Costello mailto:costello@xxxxxxxxx
<mailto:xsl-list-service@xxxxxxxxxxxxxxxxxxxxxx> wrote:

Hi Folks,

Here is a statement in my XSLT program:

<xsl:variable name="who-list" select="random:sequence($hiddennodes *
$outputnodes, 0.0, math:pow($hiddennodes, -0.5))" as="xs:double*" />

I would like to know the time required to execute each of these portions of
that statement:

(1) math:pow($hiddennodes, -0.5)

(2) $hiddennodes * $outputnodes

(3) random:sequence($hiddennodes * $outputnodes, 0.0, math:pow($hiddennodes,
-0.5))

(4) Time required to assign the variable the value

(5) Time required to ensure the value of the variable is a sequence of zero or
more xs:double values

Is there a tool that provides such fine-grain performance measurements?

/Roger

http://www.mulberrytech.com/xsl/xsl-list
http://lists.mulberrytech.com/unsub/xsl-list/673357 ()