Hi,

Bit later than planned but I've written up a post about the tuning work I 
undertook and the steps I took with some useful links:
http://joncook.github.com/blog/2012/02/12/evaluating-mark-logic-xquery-performance/

Hope it might be helpful for some.

Jon


-----Original Message-----
From: [email protected] on behalf of Geert Josten
Sent: Tue 07.02.2012 06:53
To: General MarkLogic Developer Discussion
Subject: Re: [MarkLogic Dev General] element search question
 
Hi John,

>From what I understood from it: XPath uses indexes whenever possible, but 
>depends a lot on how well the query optimizer is able to translate your XPath 
>to indexes. Slightly different expressions can give big performance 
>differences. You can investigate that with amongst others xdmp:plan. More 
>material on XPath optimizations is available in this thread: 
>http://marklogic.markmail.org/thread/swmpzuziqvfyqrlt

Apart from the optimizations, the index results often need to be filtered to 
eliminate false positives like matches a/c/b. Using cts only, and making it 
such that filtering makes no difference, xdmp:estimate returns accurate 
results..

Cheers,

Geert

Van: [email protected] 
[mailto:[email protected]] Namens John Zhong
Verzonden: dinsdag 7 februari 2012 3:53
Aan: General MarkLogic Developer Discussion
Onderwerp: [MarkLogic Dev General] element search question

Hi all,

I have xmls look like below:

<root>
<a>
<b>any text</b>
<c>
<b>any text</b>
</c>
</a>
</root>

Now, I want to search /root[a/b = 'value'] by cts constructors, for example:

cts:element-query(xs:QName("a"),
cts:element-value-query(xs:QName("b"), $value)
)

But this searches all the b element under /root/a, like /root/a/c/b, so I want 
to ask if there is any way to ignore the /root/a/c/b using cts constructors? 
Because I just want to check the /root/a/b. Or I have to use XPath like 
/root[a/b = 'value']? If so, how is the performance between cts constructors 
and XPath?

Thanks,
John


_______________________________________________
General mailing list
[email protected]
http://developer.marklogic.com/mailman/listinfo/general

Reply via email to