SQL & PHP - Which is faster mysql_num_rows() or 'select count()'?

Question

I'm just wondering which method is the most effective if I'm literally just wanting to get the number of rows in a table.

$res = mysql_query("SELECT count(*) as `number` FROM `table1`");
$count = mysql_fetch_result($res,0,'number');

or

$res = mysql_query("SELECT `ID` FROM `table1`");
$count = mysql_num_rows($res);

Anyone done any decent testing on this?

Answer 1

I guess count(1) will be even faster:

$res = mysql_query("SELECT count(1) as `number` FROM `table1`");
$count = mysql_fetch_result($res,0,'number');

Although haven't tried the proposed methods, the first makes database fetch all the records and count them in the database, the second makes database fetch a separate field for all the records and count the number of results on the server.
As a rule of thumb the less data you fetch for a particular record the less time it will take therefore I'd vote for updated first method (fetching constant for every record and counting the number of constants fetched).

Answer 2

mysql_query() transfers all result records from the MySQL into the php pcrocess before it returns (unlike mysql_unbufferd_query()). That alone would make the mysql_num_rows() version slower.

Furthermore for some engines (like MyISAM) MySQL can serve a Count(*) request from the index of the table without hitting the actual data. A SELECT * FROM foo on the other hand results in a full table scan and MySQL has to read every single dataset.

Answer 3

I don't really think any testing is needed.

Doing the COUNT in the SQL query

1) Sends only one row of data back the to client (instead of every row)

2) Lets SQL do the count for you which is likely always going to be faster then PHP.

Answer 4

Definitely the first. MySQL can usually do this by looking at an index rather than the whole table, and if you use MyISAM (the default), the row count for the table is stored in the table metadata and will be returned instantly.

Your second method will not only read the entire table into memory but also send it to the client through the network before the client counts the rows. Extremely wasteful!