I've seen a lot of people claim that MySQLi is faster than PDO and that PDO is not worth the performance hit. Unfortunately, I haven't really seen any benchmarks to back that up. So that is what I've done here.

All of the tests were done using PHP 5.3.1 (64-bit) with the MySQL Native Driver for MySQLi and PDO and MySQL 5.1.42 (64-bit) on a 2.66GHz Intel Core 2 Duo with 4GB of RAM and a 7200rpm hard drive. Each test was run 100 times. The results are the average number of milliseconds the test took plus or minus the standard deviation as a percentage of the average. Shorter bars means faster.

Each test consisted of making a connection by calling the class's constructor (MySQLi::__construct() for MySQLi and PDO::__construct() for PDO), running the queries, and then closing the connection by calling MySQLi::close() for MySQLi and setting the object to null for PDO. The times include making the connection and closing it.

These tests were very simple, but that was the point. I wanted a simple comparison that represented basic usage.

Test #1: Create a table and insert 1000 rows

Functions used:

• MySQLi::query()
• PDO::exec()

Test #2: Create a table and insert 1000 rows with prepared statements

Functions used:

• MySQLi::prepare(), MySQLi_STMT::bind_param(), and a series of calls to MySQLi_STMT::execute()
• PDO::prepare() and a series of calls to PDOStatement::execute()

With the "PDO (emulated)" test, prepared statements were emulated by setting the PDO::ATTR_EMULATE_PREPARES attribute to true using PDO::setAttribute().

Test #3: 100 selects using a "LIKE" string comparison

Functions used:

• MySQLi::query(), MySQLi_Result::fetch_array(), and MySQLi_Result::free()
• PDO::query() and PDOStatement::fetch()

Test #4: 100 selects using a "LIKE" string comparison with prepared statements

Functions used:

• MySQLi::prepare(), MySQLi_STMT::bind_param(), and a series of calls to MySQLi_STMT::execute(), MySQLi_STMT::bind_result(), MySQLi_STMT::fetch(), and MySQLi_STMT::free_result()
• PDO::prepare() and a series of calls to PDOStatement::execute() and PDOStatement::fetch()

With the "PDO (emulated)" test, prepared statements were emulated by setting the PDO::ATTR_EMULATE_PREPARES attribute to true using PDO::setAttribute().

Conclusion

For inserts, there was no significant difference between MySQLi and PDO (prepared statements or not). For selects, MySQLi was about 2.5% faster for non-prepared statements and about 6.7% faster for prepared statements. Even though PDO is considered an abstraction library, there's is very little overhead as you can see. After all, the PDO MySQL driver driver uses the same client library as MySQLi. In my opinion, the features and flexibility of PDO far outweigh the very slight performance hit. What do you think?