Question

I have a dictionary in my database which has over million records and this simple select

select * from Word where languageId = 'en' order by rand() limit 1

randomly selects one word.

The problem is that this request lasts 3-4 seconds which is very long because I have to repeat it many times.

Is there a way to achieve the same thing but much faster?

EDIT – table schema

wordId - integer, auto increment
languageId - varchar (FK), values like cs, en, de, ... 
word - varchar, word itself

Data structure example

wordId   languageId   word
--------------------------
1        cs           abatyše
...
100000   cs           zip
100001   en           aardvark
...
etc

SQL

CREATE TABLE Language (
  languageId VARCHAR(20)  NOT NULL  ,
  name VARCHAR(255)  NULL    ,
PRIMARY KEY(languageId));

CREATE TABLE Word (
  wordId INTEGER UNSIGNED  NOT NULL   AUTO_INCREMENT,
  languageId VARCHAR(20)  NOT NULL  ,
  word VARCHAR(255)  NULL    ,
PRIMARY KEY(wordId)  ,
INDEX Word_FK_Language(languageId),
  FOREIGN KEY(languageId)
    REFERENCES Language(languageId)
      ON DELETE NO ACTION
      ON UPDATE NO ACTION);

Answer 1

If you have an IDs column and the gaps between the elements are not huge (not too many elements were removed, otherwise some elements will be selected more often) then try this query

SELECT * FROM `table` 
   WHERE id >= 
      (SELECT FLOOR( MAX(id) * RAND()) FROM `table` WHERE languageId = 'en' ) 
   AND languageId = 'en'
   ORDER BY id LIMIT 1;

And look at different examples here
http://akinas.com/pages/en/blog/mysql_random_row/

ps: I just realized that it works good only without requirement for the languageId, otherwise the gaps in IDs for the same languageId could be huge.

Updated Try this one, it could be in a couple of times faster. I checked it against execution time of your query.. twice faster..

SELECT d.* FROM
  (SELECT @rn:=0 ) r, 
  (SELECT FLOOR(count(*)*RAND()) as rnd FROM `Word` WHERE languageId = 'en') t,  
  (SELECT @rn:=@rn+1 as rn, `Word`.* FROM `Word` WHERE languageId = 'en' ) d 
WHERE d.rn >= t.rnd LIMIT 1

basically it still creates some kind of continuous ids, but without sorting by them.

Last Update This one could be even faster (depends on the random number generated)

SELECT d.* FROM
  ( SELECT @rn:=@rn+1 as rn, w.*, t.rnd rnd FROM 
     (SELECT @rn:=0 ) r, 
     (SELECT FLOOR(count(*)*RAND()) rnd FROM `Word` WHERE languageId = 'en') t, 
     `Word` w 
   WHERE w.languageId = 'en' AND @rn<t.rnd
  ) d 
WHERE d.rn=d.rnd