Stalled custom field development

[Myk Melez]

Sean McAfee wrote:

>Myk Melez <myk at mozilla.org> wrote:
>  
>
>>David Miller wrote:
>>    
>>
>>>Has there been any good performance analysis on which one would 
>>>actually work faster?
>>>      
>>>
>>Yes.  I did a bunch of tests which showed FAC to be significantly 
>>faster.  See the "Custom fields schema" thread from January and February 
>>for my test results.
>>    
>>
>
>And I did a bunch of tests before that which showed just the opposite.
>Until the discrepancy can be explained, neither set of tests should be
>regarded as definitive, IMHO.
>  
>
It's not that simple.  You provided a script for generating sample data, 
some test queries you ran manually, and some results you compiled by 
hand.  I provided a script for generating sample data; a script for 
running a more comprehensive set of tests (including yours) and 
generating results automatically; and test results for two different 
machines.

I also ran the tests on a third machine and confirmed that its results 
were the same as for the first two.  My tests were more rigorous, and 
they were reproduced on several machines, so they are more likely to be 
correct than yours.

Nevertheless, note that I argue for FAD not primarily because of the 
performance implications but because it makes more sense from a design 
standpoint for the kinds of data we're storing.  I won't rehash the 
arguments here, as they're well explored in the thread earlier this year.

>One difficulty for me has been that Myk has yet to fully define his
>proposal.  Whether fields live in columns all in the same table, all in
>different tables, or divided between various tables in some fashion, and how
>the administrator would manage things in the last case, he has never stated,
>unless I've missed something.
>
It's not ambiguity.  In my proposal all three field locations are 
available, and each can be selected on a per-field basis depending on 
which location is optimal for a given field.  But we would choose a good 
default, automate the selection and updating of a location via 
administrative UI, and implement support for the locations iteratively, 
starting with the good default that most fields will use.

-myk

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.bugzilla.org/pipermail/developers/attachments/20050325/b16f8d71/attachment.html>