[rt-devel] Migration to rt3 ... final episode

mixo mixo at coza.net.za
Thu Dec 11 03:19:04 EST 2003


Jim Rowan wrote:

>My suggestions:
>
>You are aware of the issues around Apache2/mod_perl2/RH9, right?
>  
>
Right

>Do the evaluation before your production switch, not during.  
>
Thats the plan.

>Sounds
>like you have a big installation and you want to minimize down time.
>You are taking on more risk this way as well.  (What happens if you get
>stuck?)
>  
>
The process is started again, the following long weekend. The one thing 
that could get us stuck is the import
failing for what ever reason, so down should remain at the minimun 
during the next try.

>If performance matters enough to do a test, do a fair comparison.  That
>means two installs on the same hardware. 
>
This has been done already on the dual machines, and unfortunately there 
isn't another spare quad. "mysql" was faster.

> That also means taking the
>time to tune both databases.  Neither one is appropriately tuned out of
>the box.  I know that postgres performance can increase significantly; I
>suspect mysql can also.
>
What recommendations would you make for postgres (or mysql) considering 
the hardware specs?

>
>You might want to throw more than just performance into your database
>evaluation.  Do you have equivalent expertise in both?  Do you have
>other database instances already?  Do you believe them both to be equal
>in terms of features that matter?  Equal quality?
>
>If you have both machines at your disposal, consider (strongly) the
>option of using both in production; one as a dedicated database machine
>and one as the web server.  Make the bigger one the database server; the
>more memory you can give it the faster the overall system will be.  512
>is almost certainly not enough.  I think I would put the gig of memory
>in the quad machine and make it the db server...
>  
>
More  memory is a possible. I would like to add that having an 
additional rt instance means that we have backup
in case the one fails during the data import. Testing of perfomance is 
not a must, but its a good (but expensive)
exercise.





More information about the Rt-devel mailing list