Performance Tuning in Oracle for dynamic date join - oracle

Question

I have a scenario where in I have to aggregate data for a dynamic 24 hour period.
For eg: If a user selects the FROM date as Jan 05 2016 8:00 AM and TO date as Jan 10 2016 2:00 AM data in the output should be aggregated from Jan 05 2016 8:00 AM to Jan 06 2016 7:59 AM as 1 day (Jan 05 2016).
Jan 5 2016 - Jan 5 2016 8:00 AM to Jan 6 2016 7:59 AM
Jan 6 2016 - Jan 6 2016 8:00 AM to Jan 7 2016 7:59 AM
Jan 7 2016 - Jan 7 2016 8:00 AM to Jan 8 2016 7:59 AM
Jan 8 2016 - Jan 8 2016 8:00 AM to Jan 9 2016 7:59 AM
Jan 9 2016 - Jan 9 2016 8:00 AM to Jan 10 2016 2:00 AM
To achieve this, I subtracted 8 hours from the date column in the fact table and joined it to the Date Dimension. The query looks like this:
SELECT D.DAY_FMT,SUM(F.MEASURE) from FACT F
INNER JOIN DATES D ON
to_number(to_char((F.DATESTIME - 0.3333333),'YYYYMMDD')) = D.DATEID
WHERE F.DATESTIME between to_timestamp ('05-Jan-16 08.00.00.000000000 AM')
and to_timestamp ('10-Jan-16 02.00.00.000000000 AM')
GROUP BY D.DAY_FMT
Note 1: If the From Time is 06:00 AM then we would be subtracting 0.25 (days) instead of 0.3333333 (days)
Note 2: The Fact table has billions of rows.
Is there any way to improve the performance of the above query?

Answer 1

In Oracle the date and the time are stored together. You don't need to join on equality, and you don't need to wrap the date within any functions. (And why timestamps?) Having all the computations (if any are even needed) on the "right hand side" of conditions means the computations are done just once, the same for every row, instead of separately for each row.
select f.day_fmt, sum(f.measure) as some_col_name
from fact f inner join dates d
on f.datestime >= to_date('05-Jan-16 08:00:00 AM', 'dd-Mon-yy hh:mi:ss AM')
and f.datestime < to_date('10-Jan-16 02:00:00 AM', 'dd-Mon-yy hh:mi:ss AM')
group by day_fmt;
Edit: Based on further clarification from OP - suppose the data is in table "fact" - with columns day_fmt, measure, and datestime. The assignment is to aggregate (sum) measure, grouped by day_fmt and also grouped by 24-hour intervals, starting from a date-time chosen by the user and ending with a date-time chosen by the user. Solution below.
with user_input (sd, ed) as (
select to_date('05-Jan-16 08:00:00 AM', 'dd-Mon-yy hh:mi:ss AM'),
to_date('10-Jan-16 02:00:00 AM', 'dd-Mon-yy hh:mi:ss AM') from dual
),
prep (dt) as (
select (select sd from user_input) + level - 1 from dual
connect by level < (select ed - sd from user_input) + 1
union
select ed from user_input
),
dates (from_date, to_date) as (
select dt, lead(dt) over (order by dt) from prep
)
select f.day_fmt, d.from_datetime, d.to_datetime, sum(f.measure) as some_column_name
from fact f inner join dates d
on f.datestime >= d.from_datetime and f.datestime < d.to_datetime
where to_datetime is not null
group by f.day_fmt, d.from_datetime, f.to_datetime
order by f.day_fmt, d.from_datetime;
By not using function calls wrapped around f.datestime, you can take advantage of an index defined on this column of the "fact" table (an index you already have or one you can create now, to help speed up your queries).