– START SCRIPT
USE mydatabase;
CREATE TABLE `transferido_inno` (

– SOME FIELDS

PRIMARY KEY (`id`)
) ENGINE=INNODB DEFAULT CHARSET=latin1;

SET UNIQUE_CHECKS=0;
START TRANSACTION;

– SOME MULTIPLE INSERTS KIND OF
INSERT INTO transferido_inno(
SELECT * FROM transferido LIMIT 0,100000
);
INSERT INTO transferido_inno(
SELECT * FROM transferido LIMIT 100000,100000
);
–ETC.

COMMIT;
SET UNIQUE_CHECKS=1;

ALTER TABLE transferido RENAME TO transferido_myisam;

ALTER TABLE transferido_inno RENAME TO transferido;
OPTIMIZE TABLE transferido;

– END SCRIPT

It worked really good.

I started optimizing this database a month ago (my first real database challenge). This blog and your book (along with Pro MySQL) helped me very much. The site is now running reaaaaally good. The server load decreased from 90% to 20% with some tunning, new indexes and this new altered table(we needed MVCC desperately).

Thanks a lot!

Use replication, convert slave to Innodb and when switch.

I’m from Argentina, sorry for my english.

Can you give me any hint to transform all my MyISAM tables to INNOdb? The database is currently used in production, we have roughly 100 tables and 100,000 rows. The idea is to avoid the dead time of the site.

Thanks a lot for your book and your site, it’s a great help!

Thanks for the reply.

2. Yes, but say for example, MyISAM default insert speed is great, but InnoDB default insert speed is poor. (I have seen your InnoDB optimization presentation, and done the experiments)

3. This is the main point, we just afraid we will have total loss of data since we can’t even repair the table.

Some questions:

1. Is it now InnoDB still suffer from the problems of poor performance when running in SMP server?

2. If using build in default settings, seems MyISAM perform better in most suitation.

3. If InnoDB corrupt, it is 0% recoverable?

May be Innodb does not work well for your particular task (set of queries) but it is not enough to say it does not work for broad set of tasks:)

You can use GetLock (for example) to serialize data for Innodb as many people do.

Note for writes MyISAM quite often faster than Innodb but for large data sets recovery time can often be unacceptable.

I agree with you on some points. But with innodb we use repeatable-read concurrency with update if not rows been updated then insert kind of transactions. At the same time a cluster of nodes actively writing to the same tables which are read by another different cluster members. Whatever technique we tried nothing was as succesful as with MyISAM even it is doing its job not by parallel ( but inserts are faster works and/or seems like parallel ).

I am going to watch mysql webinars about scale-out write intensive applications maybe that can point a path.

You oversimplify things – in your case you get deadlocks because your data inserts are conflicting in some form – MyISAM serializes inserts and so cures deadlocks, but it is not parallel. What is better depends on the application and amount deadlocks you have.

In general the approach is simple – test what works for you, think what is important for you and make decision.