launchpad-dev team mailing list archive

[Gary Poster <gary.poster@xxxxxxxxxxxxx>]

On May 25, 2010, at 4:35 PM, Robert Collins wrote:

> On Wed, May 26, 2010 at 2:05 AM, Gary Poster <gary.poster@xxxxxxxxxxxxx> wrote:
>> Hi Björn.
>> 
>> I have a few notes.
> ...
>> - Cacheing should be an increasingly important aspect of our performance concerns, particularly of fundamental design decisions like this one.  I'd like to have some thoughts and discussion on how this plan affects cacheing (via something like memcache).  The approach to searching you discuss, for instance, should be driven from, or at the least informed by, both user interface goals and performance concerns, IMO.
> 
> I would like to raise a small red flag here. Caching is too fine
> grained a focus:  we should understand as much as we can how any new
> thing will perform and affect existing performance work. Sometimes the
> answer will be 'no widespread impact' and sometimes 'massive impact
> haaaaalp'.

...

> I'm sure that caching will remain part of our environment, but its
> important (to me) that its no more than
> part-of-checking-overall-performance.

Absolutely (except I'm not 100% I followed the last sentence exactly).

I'm not arguing that the design should incorporate cacheing, but that it should not be antagonistic to it.  This may be a fine line, but I think it is an important one to explore.  

If a design makes it difficult to express optimized queries, that's a problem.  I think Björn is trying to address that problem (though the approach seems quite challenging to implement now that I think about it--I should check out that branch).

If a design makes it difficult to cache significant chunks of the rendered result, I'm saying that may be a problem too.  Cacheing chunks of a page may be the new hotness for us, but it's certainly a fairly standard approach, and I'd prefer for our infrastructure not to work against it.

Finally, I don't think it's time to have the conversation that I am proposing until the UI and workflow have been worked through, and possibly not until after the corresponding API has been first sketched out, as Björn is doing now.

Gary