How Edmunds Got in The Fast Lane
technology.edmunds.com
technology.edmunds.com
Small sites spit out HTML from a couple of files/classes, large sites (e.g. your auto-insurance company) assemble a single HTML page using hundreds to even thousands of different files. Most large sites write their own HTML concatenator but as there is no standard or best-practice method, each one is organically grown and gets more and more complex over time. Now imagine being tasked with improving the pageload of a single webpage that calls 43 different backend services (mysql, vsam, ldap, memcache, soap etc.) across 600 different .php or .java files - you won't even know where to begin. MVC doesn't help much when you have 30 controllers calling 40 views and 20 models. The problem is there is no HTML pipeline.
Writing code to make a cube spin in 3D is easy using OpenGL. Making 100 cubes do the same isn't much more difficult either. This is because the graphics cards have a pipeline that you can fill using OpenGL commands in the right order. There is no right order for spitting out HTML, CSS, or JS and so everyone makes their own. Large sites already use some object-orientated say to echo "[div]blah[/div]" instead of hand-coding it. They just do it using their home-built solution. If there existed a standardized pipeline that different libraries, modules, and even web-servers could use, the optimizations described in the article would be nearly automatic.
HTML5, CSS3 and JS all together with common libraries like jQuery or Dojo or YUI is now pretty much resembling a GUI toolkit, and that's probably what it should have been from the start. Too bad we had to reinvent all that all over again the hard way.