Hi Tim,

Thank you for your answer.

Cordialement,

Christophe Le Roux

De : Tim Callaghan [mailto:tmcallaghan@gmail.com]
Envoyé : jeudi 2 octobre 2014 13:08
À : Christophe Le Roux
Cc : MARK CALLAGHAN; Tim Callaghan; Maria Discuss; Jan Lindström
Objet : Re: [Maria-discuss] MariaDB,TokuDB and FusionIO

One drawback to mixed engines inside single transactions is the need for XA prepare (and extra fsyncs).

On Wed, Oct 1, 2014 at 3:36 AM, Christophe Le Roux <christophe.le.roux@fr.clara.net> wrote:

Hi Mark,

Tim Callaghan ( same lastname, your brother ? ;-) )from Tokutek told me similar things :

“As for benefit, that's to be determined but since we don't have to double write like InnoDB that isn't important to us, plus we are extremely write optimized.“

If we have FusionIO device and mixed tables (innodb, tokudb), is there any benefit (or drawback ?) to use NVMFS with TokuDB table ?

Cordialement,

Christophe Le Roux

De : MARK CALLAGHAN [mailto:mdcallag@gmail.com]
Envoyé : mardi 30 septembre 2014 18:23
À : Christophe Le Roux; Tim Callaghan
Cc : Jan Lindström; Maria Discuss

Objet : Re: [Maria-discuss] MariaDB,TokuDB and FusionIO

AFAIK:
* TokuDB does large random writes, not small ones
* TokuDB doesn't do update in place

So it doesn't need something like the doublewrite buffer in InnoDB. I am not sure whether they might benefit from atomic writes. They would benefit from access to a few nice/fast FusionIO devices.

On Tue, Sep 30, 2014 at 2:12 AM, Christophe Le Roux <christophe.le.roux@fr.clara.net> wrote:

Hi Jan,

Thank you for your answer.

I am writing to the TokuDB community.

I hope for good news J

Cordialement,

De : Jan Lindström [mailto:jan.lindstrom@mariadb.com]
Envoyé : mardi 30 septembre 2014 10:38
À : Christophe Le Roux; Maria Discuss
Objet : Re: [Maria-discuss] MariaDB,TokuDB and FusionIO

Hi,

On Tue, Sep 30, 2014 at 11:27 AM, Christophe Le Roux <christophe.le.roux@fr.clara.net> wrote:

About NVM compression, is it a hardware compression doing by the FusionIO controller or is it a software compression doing by the CPU of the server ?

Compression is done by CPU in the MariaDB server (InnoDB/XtraDB storage engine).

About Atomic Write, I understand there is missing code to have this feature to the TokuDB engine. Should I ask to the Mariadb dev team or the Tokutek dev team (or SanDisk/FusionIO dev team ?) to add it (maybe it’s technically no doable ?)?

I would start from Tokutek dev team, I do not know fully how database tables are stored on Tokutek, but I do not see immediately a reason why atomic writes would not be possible.

Do you think this can boost performances for huge TokuDB tables ?

Atomic will boost performance but only if database is stored on device that supports atomic writes.

R: Jan

_______________________________________________
Mailing list: https://launchpad.net/~maria-discuss
Post to : maria-discuss@lists.launchpad.net
Unsubscribe : https://launchpad.net/~maria-discuss
More help : https://help.launchpad.net/ListHelp

--
Mark Callaghan
mdcallag@gmail.com