How Top And Set Rowcount Actually Works

I'm coming from Oracle world and my knowledge about SQL Server is quite limited so I apriori apologize for probably stupid questions

DB version is SQL Server 2005.
The business scenario is - there is search form with many criteria. User may enter very unrestrictive criteria matching probably millions of rows. To prevent that we'd like to show him no more than N rows (N ~200). Any rows matching criteria are good enough, however these FOUND rows we'd like to sort. I do not want to find all potentially X million rows, then sort them and only then show forst N rows.
So in Oracle I know how to do that. I just find N rows in subquery, and then in outer query sort them. So I avoid to find all rows and then sorting millions of rows.
Here is an example:



Code Block
SQL> create table t (id number, data varchar2
Table created.
SQL> insert into t values (1, 'aaa');
1 row created.
SQL> insert into t values (2, 'bbb');
1 row created.
SQL> insert into t values (3, 'ccc');
1 row created.
SQL> select * from (
2 select * from t where rownum <=2)
3 ;
ID DATA
---------- --------------------
1 aaa
2 bbb
SQL> ed
Wrote file afiedt.buf
1 select * from (
2 select * from t where rownum <=2)
3* order by data desc
SQL> /
ID DATA
---------- --------------------
2 bbb
1 aaa





However how can I avoid sort of potential big result in SQL Server? I've searched google but unfortunately found nothing.
I've tried to use both TOP and SET rowcount without success i.e. from these examples I assume that DB will find ALL rows matching where clause then sort them keeping only first N. It seems that order by clause in outer query is pushed into inner query both for top and set rowcount.




Code Block
create table t (id integer, data varchar(20));
insert into t values (1, 'aaa');
insert into t values (2, 'bbb');
insert into t values (3, 'ccc');

select * from (
select top 2 * from t) as q


1 aaa
2 bbb


select * from (
select top 2 * from t) as q
order by data desc

3 ccc
2 bbb


set rowcount 2
select * from (
select * from t) as q


1 aaa
2 bbb


set rowcount 2
select * from (
select * from t) as q
order by data desc

3 ccc
2 bbb







And I'd like to avoid sort because of two reasons:
1) I predict that generally finding all rows will be much more costly than finding just any no more than N
2) Sorting all found rows also probably will be slower than just N (however DB has to find only first N rows, so not ALL rows should be sorted/kept sorted)


Are my concerns reasonable? If yes what can I do to just find N rows and sort only these?

TIA, Gints