Structs As Query Indexes, Speed, And Rick Osborne
For those of you who follow my blog, you will know that Rick Osborne is the guy who comes in after I explain things and makes killer suggestions about how they can be done better. I recently gave a case study of how ColdFusion code can be optimized. It involved using the IndexOf() method of the ColdFusion query column object. Rick came in and suggested that using a Struct to create your own query-index would perform faster. As I am a man who likes to learn by doing, I thought I would put this to the test.
To test this, I query from a web statistics program. I am getting information from two tables: web_stats_hit and web_stats_session. Each "hit" in the hits table will have a session id in the session table. The session table also have a UUID column "session_id".
As I am trying to test the merging of two data sources, I am going to hit each table individually and then try to update one with matching values from the other. First I am grabbing the two different data sets:
<!--- Query for web hits. --->
<cfquery name="qHit" datasource="...">
SELECT
h.id,
h.date_created,
h.web_stats_session_id,
(
''
) AS session_id
FROM
web_stats_hit h
</cfquery>
<!--- Query for web sessions. --->
<cfquery name="qSession" datasource="...">
SELECT
s.id,
s.session_id
FROM
web_stats_session s
</cfquery>
As I tried to explain earlier, for every qHit.web_stats_session_id, there is a matching session, such that for some combo, qHit.web_stats_session_id == qSession.id. And just to get an idea of the amount of data we are talking about:
qHit: 52,290 records
qSession: 34,753 records
That's a LOT of data to go through. Let's get our "test" on:
<cftimer label="IndexOf() Methodology" type="outline">
<!--- Loop over the hit query. --->
<cfloop query="qHit">
<!---
We want to find a matching session_id based
on the session. Get index of matching row.
--->
<cfset intIndex = qSession[ "id" ].IndexOf(
JavaCast( "int", qHit.web_stats_session_id )
) />
<!--- Add one to index (to be ColdFusion friendly). --->
<cfset intIndex = (intIndex + 1) />
<!--- Check to see if we have an index. --->
<cfif intIndex>
<!--- We found the match, update the row. --->
<cfset qHit[ "session_id" ][ qHit.CurrentRow ] =
qSession[ "session_id" ][ intIndex ]
/>
</cfif>
</cfloop>
</cftimer>
<cftimer label="Struct Index Methodology" type="outline">
<!--- Create a session look up table. --->
<cfset objSessionLookUp = StructNew() />
<!---
Loop over session and set index rows. We will be using
the id column of the session as the key and the
session_id as the value. This creates our very own,
in-memory index of the qSession query based on ID.
--->
<cfloop query="qSession">
<!--- Index this value. --->
<cfset objSessionLookUp[ qSession.id ] = qSession.CurrentRow />
</cfloop>
<!--- Loop over the hit query. --->
<cfloop query="qHit">
<!---
Check to see if the session key exists. If it
does, then we found a match.
--->
<cfif StructKeyExists(
objSessionLookUp,
qHit.web_stats_session_id
)>
<!--- Update the session based on the struct-index. --->
<cfset qHit[ "session_id" ][ qHit.CurrentRow ] =
objSessionLookUp[ qHit.web_stats_session_id ]
/>
</cfif>
</cfloop>
</cftimer>
It turns out Rick was absolute correct. The struct index performs MUCH faster. Here are the stats:
IndexOf() Methodology : 402,037 ms
Struct Index Methodology : 33,403 ms
Some quick math will show you that the struct index method performs in 8% of the time that the IndexOf() methodology does. EIGHT PERCENT! Nuts. I guess the only downside is that you can potentially create a HUGE in-memory structure; but it's only temporary.
Nicely done Rick!
Want to use code from this post? Check out the license.
Reader Comments
I figured you'd appreciate this:
http://rickosborne.org/images/screenshots/wtg-ben.png
That graph represents the last week's traffic to my site. This site has accounted for 3% of the traffic to my site, behind only Google and MXNA.
Rick,
That's awesome :) Glad to be sending people to you. You have good information to share.