| 1 |
Ramon van Alteren wrote: |
| 2 |
> kashani wrote: |
| 3 |
>> Ramon van Alteren wrote: |
| 4 |
>> |
| 5 |
>>>> did NTPL help you guys? |
| 6 |
>>>> |
| 7 |
>>> |
| 8 |
>>> Strangely it actually hurt our mysql-performance. Although mysql AB |
| 9 |
>>> recommends it. Haven't done any recent testing however. |
| 10 |
>> |
| 11 |
>> That's really odd. I've never seen Mysql do anything, but get much |
| 12 |
>> much faster with NTPL at least with our work loads. Web servers, |
| 13 |
>> mostly selects, PHP, 600-1000 connections to each of the db server. |
| 14 |
>> Load went from 1.5 to .3 on our dual proc boxes when we moved from |
| 15 |
>> 2.4 to 2.6 w/NTPL. I suspect it's the number of connections we have |
| 16 |
>> that caused most of the benefit in our case. |
| 17 |
> |
| 18 |
> Let's start with "which version ?" |
| 19 |
> |
| 20 |
> I'd have to look through all testing docs for the period but I |
| 21 |
> remember that we tested NTPL with 4.0 and found it hurt our performance. |
| 22 |
NTPL was un-supported by MySQL on 4.0, see [1], and support started not |
| 23 |
much more than one year ago (prior patching the sources was needed) |
| 24 |
> |
| 25 |
> Similar work-load: webservers, reading from slaves, writing only to |
| 26 |
> the replication-master, PHP |
| 27 |
> Maybe different sizing? We have a very large database spanning well |
| 28 |
> over 35Gb now, with some extremely large tables in it. Most dbservers |
| 29 |
> are IO-bound not CPU-bound. |
| 30 |
then switching thread implementation model would make no difference ;-) |
| 31 |
|
| 32 |
[1] http://bugs.mysql.com/bug.php?id=19785 |
| 33 |
|
| 34 |
-- |
| 35 |
gentoo-server@g.o mailing list |
Archives