I wonder whether you might not see an improvement if you accumulate all
the values you are searching for and find them all at once:
let $code := (1, 2, 3, ...)
let $items := //it...@code=$code]
not sure whether that fits your application well, but it might perform
better. Sometimes I also find that for reasons that are obscure to me,
I get better performance using the cts:search/cts:query combination even
for expressions that look like they could really be optimized for me - I
guess optimizing is a black art, although it seems that ML does keep
improving it. If you do that, the element-value-query is what you want...
-Mike
Lee, David wrote:
thanks, I actually do use eq not = ...
-----Original Message-----
From: general-boun...@developer.marklogic.com
[mailto:general-boun...@developer.marklogic.com] On Behalf Of Micah
Dubinko
Sent: Tuesday, November 10, 2009 6:41 PM
To: General Mark Logic Developer Discussion
Subject: Re: [MarkLogic Dev General] Optimize exact match searches
onattributes
David,
I'm guessing the optimizer already does this, but just in case, try
using the atomic comparison operator eq instead of =. (The value in
$code will never be a sequence, right?)
$root/tree/item[ @code eq $code ]
Even though it goes against conventional wisdom, also try $root//item
[ ... ]
You can use xdmp:elapsed-time() to get decent resolution timing data
to experiment with what works the fastest.
-m
On Nov 10, 2009, at 3:15 PM, Lee, David wrote:
Suppose I have an XML document with about 50,000 of these small
elements
<item name="Peritoneum" parent="A01.047.025"
code="A01.047.025.600"/>
And an Xquery that does a LOT of this kind of search
let $item = $root/tree/item[ @code = $code ]
or
let $item = $root/tree/item[ @parent = $code ]
These are exact string matches, no fancy terming or wildcards or
anything.
Any suggestions on the basic area to look for optimizing this kind
of access ? Either using xquery or a setting on the server ?
Thanks for any suggestions. So far I've only been able to find
things tailored for full text searching (1 letter, 2 letter, word
searches etc).
I would think direct attribute=value searches could be tagged for
efficient indexing. Maybe they are already ?
But its starting to slow down when I do about 20 of these in an
xquery.
One thought I had was to preprocess the data into more of a tree
structure instead of the flat structure it is now ...
<item >
<item>
<item>
</item>
</item>
then I can use ancestor axis to search up the tree instead of a flat
search ... But if these were already indexed it "should" be a near
instant access as a flat structure (???)
----------------------------------------
David A. Lee
Senior Principal Software Engineer
Epocrates, Inc.
d...@epocrates.com
812-482-5224
_______________________________________________
General mailing list
General@developer.marklogic.com
http://xqzone.com/mailman/listinfo/general
_______________________________________________
General mailing list
General@developer.marklogic.com
http://xqzone.com/mailman/listinfo/general
_______________________________________________
General mailing list
General@developer.marklogic.com
http://xqzone.com/mailman/listinfo/general
_______________________________________________
General mailing list
General@developer.marklogic.com
http://xqzone.com/mailman/listinfo/general
