Scalability issues for dummies
alwaysnewmistakes.wordpress.com
alwaysnewmistakes.wordpress.com
I have mentioned it three or four times before but I'll say it again: the key takeaway from the YSlow presentations is that fixing the frontend is a science, not black magic and that user-perceptible load times are dominated by front-end factors.
Not your database! Not your application logic! Not your web framework! Not your idiom for string concatenation!
If you have three CSS files, condensing them to 1 CSS file is as close to zero technical risk as any intervention in the history of engineering, will take you far under a man-day to implement a process to correct this defect, and will provide instant, lasting, guaranteed performance gains everywhere you were using those three files.
If you don't gzip your Javascript, CSS, and HTML yet...
If you have spent weeks thinking about how to use memcached and not so much as a minute discussing expires headers...
I do agree with you that a lot of what users experience as slowness IS front-end stuff. BUT there is also a lot of issues with the backend not working as expected. Lets say, front-end optimization is like shooting the low hanging fruit :) Nevertheless you're right I should have commented it somewhere.
It is true that user-perceptible load times are dominated by front-end factors. However, whether the user perceives anything at all -- or if, instead, the web site has collapsed under load -- is dominated by back-end scalability.
Scalability is far more about keeping things running than it is about making things fast. People should pay attention to both front-end and back-end issues -- a website which is up but loads very slowly isn't very useful, but neither is a website which loads very fast but goes offline whenever it gets a lot of traffic.
The need to state these facts is further evidence that this is an arts and crafts industry (and not scientific). Imagine physicist saying "Inertia should not be confused with momentum".
I suppose you could arbitrarily redefine scalability such that problems caused by scale which did not cause the server to crash were not scalability problems, but your redefinition would not contribute to professionalizing the field of web development.
"professionalizing the field of web development"
I think this touches on a key differentiator of our chosen profession. Nearly every other (industrial) professional field (with the exception of Architecture, which shares many of the characteristics of Software "engineering" in sharing a hazy boundary between 'art' and 'science') has a tiered body of workers that rests on a fairly commoditized implementation base. (Consider your precious laptop. Put together by workers with high-school education.)
I don't see how we can "professionalize" [software] development (in terms of actual producers of ubiquitous software artifacts) unless we have programs writing programs.
You can have a website that takes 3 minutes to load yet still be scalable. As long as it takes 3 minutes to load when you have a single user, 3 minutes at 1000 users and 3 mintes at 1 million you have a scalable website. Scalability is all about keeping the same (or at least close) performance as your load increases.
If you want to know more about scalability I suggest you read Theo Schlossnagle's excellent Scalable Internet Architectures book.
On the other hand, it'd be pretty strange if a blog post had to mention the definition for encapsulation. That stuff seems pretty well-known and agreed.
If the server is crashing or only intermittently serving pages because of load there is nothing front end optimization can do about it.
I guess sometimes you also have to weigh up just adding some extra hardware to smooth a transition to a new back end, or even to buy some more time with the existing one.