Re: MySQL performance problem on MacOS X

Hi Guys,

This worried me so I did some internet research and the results are a little more reassuring:

1. Anandtech screwed up by using a crippled compiler for the G5 test and have misunderstandings of process / thread handling in OS X. (see comments for anandtech article and http://ridiculousfish.com/blog/?p=17)

2. Using Apache2 with it's better multiprocessor capabilities will help.

3. Some say that the performance problem may be related to the built-in F_FULLFSYNC fcntl which will ask the drive to flush all of its buffered data to disk. This has been questioned and at the OS level it is always on. it is possible to turn this off if the app allows it. However this is a tradeoff between speed and data integrity. (http://lists.apple.com/archives/darwin-dev/2005/Feb/msg00072.html) Written by the dev who wrote BeOS and now works for Apple - excellent thread.

4. On the Mac OS X Server list many people ran tests and got much better results than the anandtech report - some on G4s! Dr Rolf Jansen pointed out that setting: /usr/sbin/sysctl -w net.inet.tcp.delayed_ack=0 overcomes problems with cheap NICs with too small transfer buffers. He suggests that Anandtech "simply had very low network performance from their PC-Box (possibly with a noname NIC with a small transfer buffer) to the Dual G5 Server."

5. The inimitable Dan Shoop (also on OS X server list) acidly points out that MySql isn't enterprise level and that PostGreSQL is muuuuuch better. He reminds us that MySql "strongly favours Linux as a platform" and "performance on OS X is tempered with a far greater level of data integrity than you get on Linux. For most ppl interested in databases raw performance isn't as important as safe operations. Data is more valuable than speed."

6. The results - OS X may be a slower, but not 10x as reported by anandtech and your data integrity is better. Some test run a year ago by PC magazine give an entirely different picture: http://www.pcmag.com/article2/0,1759,1637655,00.asp

7. I'd say the jury is out but I'm interested in Dr Rolf's sysctl suggestion. Anyone who has experienced poor mysql performance tried this??

cheers
adam.

At 12:30 PM -0400 6/6/05, Ken Whitcomb wrote:

So if I understand your post and the link correctly, the choice is between performance and data integrity. If I were in an environment where I knew that my backup power would not fail me, I might consider the speed option. On the other hand, regardless of the OS, I'm not sure that I want my customer's db spending part of it's time in limbo in a disk cache. I think that any OS other than MOSX that runs MySQL should consider some sort of script that'll flush the disk cache on some reasonable period.

Ken Whitcomb

On Jun 6, 2005, at 11:26 AM, David Kazias wrote:
People who actually know what's going on have placed the problem right at
the feet of the fact that under OS X MySQL defaults to performing a full
flush of the hard disk cache to ensure that a transaction has been written
to disk, whereas under Linux it doesn't.
http://lists.apple.com/archives/darwin-dev/2005/Feb/msg00072.html

Supposedly you can turn this off, but I've not bothered to look for how.
---------
Tenon Intersystems' iTools Mailing List
To unsubscribe: send mail to itools-request@xxxxxxxxxxxxxxx
with the body only containing: unsubscribe
Find the searchable mailing list archives at:
http://www.tenon.com/lists/html/iTools/

--------- Tenon Intersystems' iTools Mailing List To unsubscribe: send mail to itools-request@xxxxxxxxxxxxxxx with the body only containing: unsubscribe Find the searchable mailing list archives at: http://www.tenon.com/lists/html/iTools/