Question

I have a large table I have been playing with the query analyzer and looking for the best way to do this.

The table is like this:

name         rows        reserved     data         index_size   unused
table_name   110980132   7802944 KB   6119784 KB   1679320 KB   3840 KB

And has these columns:

 ID int, time_stamp datetime, value1 float, value2 float, value3 float.... 

These time_stamps are dates with times. I need to find an easy way, without storing anything, to be able to get just the date portions for the table. Eventually, I may need to know just the day + hour part (and not the whole time portion). At the moment, I just need to know what the last 30 days we had data for are (sometimes days are missing at this point, this question/query will ultimately not just be looking for the last x days, but all the days, or whatever).

What is the best way to do this considering performance and time? I’ve played with group by, distinct, top x, rank(), temp tables, views… some things are better than others but nothing I am doing seems to be great.

Ideas? Thanks!

Answer 1

If you’re open to using T-SQL batches, instead of a single query, then you can make use of indexes like this:

create table #tmp (date datetime primary key clustered);
declare @pivot datetime;
  insert #tmp
  select TOP(1) datediff(d,0,time_stamp)
    from tbl
order by time_stamp desc;
while @@rowcount > 0 and (select count(*) from #tmp) < 30
begin
      insert #tmp
      select TOP(1) datediff(d,0,time_stamp)
        from tbl
       where time_stamp < (select min(date) from #tmp)
    order by time_stamp desc;
end;

All this requires of you is a good index on time_stamp, and it will perform exactly 30 seeks (or less) on that index. Very surgical and quick. I threw it up as a concept, so obviously the 2 scalar subqueries in there can be easily optimised.