Question

How can I compress / aggregate / group a table with events dynamically over time.
I have a table with values and time of occurrence.

Something like this:

value_col   time_col
3         | 2011-02-16 22:21:05.250
2         | 2011-02-16 21:21:06.170
15        | 2011-02-16 21:21:05.250

I need to aggregate the values by a given time span (e.g. hourly) starting from the first row (latest event). So in this example I want to end up with two rows for hourly aggregation.

5
15

So if a new value comes in:

value_col   time_col
6         | 2011-02-16 23:21:05.247
3         | 2011-02-16 22:21:05.250
2         | 2011-02-16 21:21:06.170
15        | 2011-02-16 21:21:05.250

If I would run that query again I want to end up with:

9
17

It should be easy to change the time span in the query. For example compress over the last 30 seconds, past 6 hours, past 24 hours , etc.. How can I do that in oracle and MS SQL?

Answer 1

Thanks to the previous answers I got the idea on how to fulfill all the requirements.

For each record I calculate the time difference to the latest record in milliseconds (or seconds, depending on resolution). I then modulo the difference with the time span that I am currently interested in (e.g. 3600 sec = 1 h).
Then I add that value to the time_col of the same record and group over that.

Create table:

CREATE TABLE [dbo].[test_table](
    [value_col] [int] NOT NULL,
    [time_col] [datetime] NOT NULL
) ON [PRIMARY]
GO
INSERT [dbo].[test_table] ([value_col], [time_col]) VALUES (3, CAST(0x00009E8C01705737 AS DateTime))
INSERT [dbo].[test_table] ([value_col], [time_col]) VALUES (2, CAST(0x00009E8C015FDD8B AS DateTime))
INSERT [dbo].[test_table] ([value_col], [time_col]) VALUES (15, CAST(0x00009E8C015FDC77 AS DateTime))
INSERT [dbo].[test_table] ([value_col], [time_col]) VALUES (6, CAST(0x00009E8C0180D1F6 AS DateTime))

Solution for SQL:

SELECT SUM(value_col) AS s_val, aggregation_time FROM 
 (SELECT value_col, time_col, 
  DATEADD(millisecond,DATEDIFF(millisecond,time_col,(SELECT MAX(time_col) 
  FROM test_table)) % (3600 * 1000), time_col) AS aggregation_time 
 FROM test_table)
GROUP BY aggregation_time
ORDER BY aggregation_time DESC

Solution for Oracle:

SELECT SUM(value_col) as s_val, aggregation_time FROM
 (SELECT value_col, time_col + 
  (MOD(ROUND(((CAST((SELECT MAX(time_col) FROM test_table) AS DATE ) - 
  CAST(time_col AS DATE ))*86400),0),3600))/86400 as aggregation_time
  FROM test_table l)     
GROUP BY aggregation_time
ORDER BY aggregation_time DESC

If I want to aggregate over the last 2 h I just change 3600 to 7200 seconds.

The result is:

9   2011-02-16 23:21:05.247
17  2011-02-16 22:21:05.247