Today I read a nice article , which give a new design for mysql database . In this article , Author (Mr.lou) uses OneSQL (a branch of MySQL) for the whole test.

you can read from :

This article explain how to use mysql in seckilling scenario (especially for E-Commerce commany )

If you want to use transaction in RDBMS database, you must manual start transaction and commit in the end of your operation.
So,commit will be cost of the total process, round-trip cost will be the most important cost as author said

(For simple update operations, without regard to conflict of IO and locks, a statement execution time is about 0.1ms, network delay under normal conditions is 0.4-0.8ms, that the waiting time is longer than the transaction commit time to really notice the SQL execution several times, showing that the network is a single-line transaction update performance killer)

How to avoid this ? We can start a transaction to manually control SQL commit. When autocommit=1 was set on MySQL server implicitly commit will happen after every SQLs.

An explicit transaction is when statement(s) are wrapped within an explicitly defined transaction code block - for MySQL, that's START TRANSACTION. It also requires an explicitly made COMMIT or ROLLBACK statement at the end of the transaction. Nested transactions is beyond the scope of this topic.

Implicit transactions are slightly different from explicit ones. Implicit transactions do not require explicity defining a transaction. However, like explicit transactions they require a COMMIT or ROLLBACK statement to be supplied.

Using OneSQL ,Commit operation can be done on MySQL server side to avoid network round-trip. In total logical process, everything will be controlled by yourself .

In seckilling scenario, this optimization will give a tremendous performance promotion.

Actually ,we can also use Redis or memcached for this situation, Redis CAS feature witch MULTI in serial process will be a good choice.

Best practice - try to offload request before database. Order creation should be asynchronous (put operations in a serial async queue is the best choice)