Returning random rows from mysql database without using rand()

Question

I would like to be able to pull back 15 or so records from a database. I've seen that using WHERE id = rand() can cause performance issues as my database gets larger. All solutions I've seen are geared towards selecting a single random record. I would like to get multiples.

Does anyone know of an efficient way to do this for large databases?

edit:

Further Edit and Testing:

I made a fairly simple table, on a new database using MyISAM. I gave this 3 fields: autokey (unsigned auto number key) bigdata (a large blob) and somemore (a medium int). I then applied random data to the table and ran a series of queries using Navicat. Here are the results:

Query 1: select * from test order by rand() limit 15

 Query 2: select * 
          from 
      test 
          join 
      (select round(rand()*(select max(autokey) from test)) as val from test limit 15)                                           as rnd
      on 
          rnd.val=test.autokey;`

(I tried both select and select distinct and it made no discernible difference)

and:

Query 3 (I only ran this on the second test):
SELECT  *
    FROM    (
    SELECT  @cnt := COUNT(*) + 1,
            @lim := 10
    FROM    test
    ) vars
    STRAIGHT_JOIN
    (
    SELECT  r.*,
            @lim := @lim - 1
    FROM    test r
    WHERE   (@cnt := @cnt - 1)
            AND RAND(20090301) < @lim / @cnt
    ) i

ROWS:            QUERY 1:               QUERY 2:         QUERY 3:
2,060,922          2.977s                 0.002s            N/A

3,043,406          5.334s                 0.001s            1.260     

I would like to do more rows so I can see how query 3 scales, but at the moment, it seems as though the clear winner is query 2.

Before I wrap up this testing and declare an answer, and while I have all this data and the test environment set up, can anyone recommend any further testing?

Answer 1

Try:

select * from table order by rand() limit 15

Another (and possibly more efficient way) would be to join against a set of random values. This should work, if there's some contiguous integer key in the table. Here is how I would do it in postgres (My MySQL is a bit rusty)

select * from table join 
   (select (random()*maxid)::integer as val from generate_series(1,15)) as rnd
   on rand.val=table.id;

where maxid is the highest id in table. If id has an index, then this would mean only 15 index lookup, so its very fast.

UPDATE:

Looks like there no such thing as generate_series in MySQL. My fault. We don't need it actually:

select * 
from 
 table 
join 
 -- this just returns 15 random numbers. 
 -- I need `table` here only to produce rows for rand()
 (select round(rand()*(select max(id) from table)) as val from table limit 15) as rnd
on 
 rnd.val=table.id;

P.S. If I don't want duplicates returned, I can use (select distinct [...]) in the random generator expression.