[ale] mysql advice

Stephen Benjamin skbenja at gmail.com
Fri Jun 19 18:26:43 EDT 2009


I think the easiest solution here is to use mysqlslap, here's some info on
its use:  http://www.jameslittle.me.uk/mysqlslap-for-mysql-50/.  You tell it
how many iterations, how many connections to make, and what SQL to submit
(or it can generate some of its own).

But... 200 daemons entirely depends on what SQL they're submitting to the
server and the hardware you're running it on.  If they're doing lots of
INSERTS and some simple SELECTS, I think it would probably be fine on
relatively modern hardware.  If they're all submitting complicated JOINS
then you might have a problem.   Also, your DB engine comes into some play
here.  If you're willing to sacrifice transactionalness you can go with
MyISAM which is very, very fast and you might see some performance gains
using it.  It's  not crash-safe.  Maria is an engine written by Monty
(former CEO of MySQL AB) which is MyISAM but crash-safe but I haven't used
it.

Also, if the end result is that 1 MySQL server just isn't going to be able
to handle your load, then there are options.  Master-master (dual or more)
mysql servers is a possibility, depending on the consistency of data you
need (can you handle retrieving data that might be a few minutes old)?
Also, MySQL Cluster is an option in limited circumstances -- it doesn't
handle complicated stuff well, but it does handle LOTS and writes and LOTS
of reads as long as you're not using FK's or JOINs.  And finally, commercial
solutions like uni/cluster are available.


- Steve


On Thu, Jun 18, 2009 at 9:20 AM, Atlanta Geek <atlantageek at gmail.com> wrote:

> We have an application that has about 20 daemons on a single server
> writing data to a mysql database.
> A new requirement has popped up that suggests we need 200 daemons on
> multiple servers writing data to a mysql database.
>
> Does anyone know if there is a client limit to mysql. Currently these
> daemons will be writing to the same tables. Any suggestions.  This
> situation is difficult to simulate in the lab unless we write
> speciallized code.  Any suggestions from experience would be
> appreciated.
>
>
> --
> http://www.atlantageek.com
> _______________________________________________
> Ale mailing list
> Ale at ale.org
> http://mail.ale.org/mailman/listinfo/ale
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.ale.org/pipermail/ale/attachments/20090619/f4a3f3d0/attachment.html 


More information about the Ale mailing list