views:

281

answers:

2

Basically I want is to aggregate some values in a table according to a timespan.

What I do is, I take snapshots of a system every 15 minutes and I want to be able to draw some graph over a long period. Since the graphs get really confusing if too many points are shown (besides getting really slow to render) I want to reduce the number of points by aggregating multiple points into a single point by averaging over them.

For this I'd have to be able to group by buckets that can be defined by me (daily, weekly, monthly, yearly, ...) but so far all my experiments had no luck at all.

Is there some trick I can apply to do so?

+4  A: 

I had a similar question: collating-stats-into-time-chunks and had it answered very well. In essence, the answer was:

Perhaps you can use the DATE_FORMAT() function, and grouping. Here's an example, hopefully you can adapt to your precise needs.

SELECT
    DATE_FORMAT( time, "%H:%i" ),
    SUM( bytesIn ),
    SUM( butesOut )
FROM
    stats
WHERE
    time BETWEEN <start> AND <end>
GROUP BY
    DATE_FORMAT( time, "%H:%i" )

If your time window covers more than one day, and you use the example format, data from different days will be aggregated into the 'hour-of-day' buckets. If the raw data don't fall exactly on the hour, you can smooth that out by using "%H:00".

Thanks be to martin clayton for the answer he provided me.

cmroanirgo
Does this scale? My problem is that this will produce several millions of entries after a year.
cdecker
I can't see why it wouldn't. Obviously doing any kind of time conversions is slow, but most of the time will be spent in the aggregation of data itself, which is unavoidable in your case.
cmroanirgo
A: 

It's easy to truncate times to the last 15 minutes (for example), by doing something like:

SELECT dateadd(minute, datediff(minute, '20000101', yourDateTimeField) / 15 * 15, '20000101') AS the15minuteBlock, COUNT(*) as Cnt
FROM yourTable
GROUP BY dateadd(minute, datediff(minute, '20000101', yourDateTimeField) / 15 * 15, '20000101');

Use similar truncation methods to group by hour, week, whatever.

You could always wrap it up in a CASE statement to handle multiple methods, using:

GROUP BY CASE @option WHEN 'week' THEN dateadd(week, .....
Rob Farley