[TYPO3-core] Re: Memory system woes

[Mathias Schreiber]

Quote: Michael Stucki (stucki) wrote on Wed, 13 February 2013 00:45
----------------------------------------------------

> The learning should be:
> - Code cleanup will improve the maintainability of the overall code
> and should be encouraged

> - Performance should not be removed without having a follow-up patch
> at hand that improves the situation again

This is really really NO flame or something.
I don't understand, why someone should be allowed to decrease performance and then be obligated to have a "follow-up" at hand.
Why not simply have these two steps in one?
Like "oh, I messed up parsetimes here or memory consumption there, I will not commit until I fixed it".

I see the following benefits:
a) Less frustration (because our experience shows that folks WILL mess up what you built without giving a f**k about performance).
Best example is the "do not load in FE flag" which has been removed.
The removal improved nothing but restored the previous (worse) state of RAM usage.
At least to me this clearly signals "I have no idea what this does, let's remove it".
I, personally am frustrated by this and I know so from Rupi and Joey (which besides Stucki is the entire performance team... go figure).

b) Better communication:
No one needs to be pissed off because the people CHANGING things will also keep track of what they might break performancewise.

c) Less patches:
Who doesn't want to work less, hu? :)

d) Better outbound communication:
Marketing folks can hook up to the claim that "performance matters to the TYPO3 team".
Benefit for everybody.

It all comes down to responsibility.
And changing things without benchmarking them afterwards is simply irresponsible.
And this is what pisses Dimitry off.
Simple as that.

A general sidenote on performance:
Website performance is like a First person shooter - the framerate cannot be too high.
Means: a website can never be "too fast".

About the whole "we have more RAM today than 10 years ago blabla":
All I say is: horizontal scaling.
Cloud, Blades etc.
Currently IT strategies go towards using LESS RAM in order to keep the datacenters cheaper.
We maintain a high load application which may only be hosted on machines with 1 GB of RAM.
Why?
Because the app requires us to scale with the amount of machines - not the amount of RAM in a single box.

Stucki asked the other "sides" (why do we have sides anyways? Aren't we all on the same track here?) to try and understand each other.
Right now (at least for the last 7-8 years) we have the "I don't care about speed" faction doing as they like while the "umm... we have a problem here" faction are the ones to solve the problems other people introduced.
I am pretty sure, a lot of all the hassle could be avoided by faction "refactor" paying attention to faction "it's slower than before" just ONCE.

You all know my home adress in order to send me pipe-bombs :)
Cheers
mathias