> Fair enough. I shouldn't have said "lousy performance of the > framework itself" when I should have included the application. If the > application's page computations are so lengthy, then they too need > speeding up. > > We've got a situation where some big sites (Slashdot, Wikipedia) have > a lot of cached static pages for non-logged-in users (they all see the > same thing), but any user who is logged in sees a version customized > by their preferences, that's usually not cached. So there's a > perverse incentive to not log in, since you see the static page > faster. > > I'd really like to get hold of a big active blog or BBS server to > profile it. It's been puzzling me for years what makes them so slow. > They just paste user-contributed content together with HTML from > templates, so you'd think it shouldn't be too complicated.
Most of the time, that means fetching data from the DB, which means context switches and network transfer. I'm a developer on a large java-driven application that deals with books. The app benefits hugely from caching - the whole object model is rather elaborated, and fetching it into memory (including images stored as blobs), and serializing it takes a couple of seconds. Per user! But just grabbing it from a disk as html snippet speeds up the app tremendously. Additionally, commercial sites often are composed by a rather large number of different parts. Teasers, lists of e.g. thematically related content and so on. And if you have lots of comparably large objects that are very diverse, a larger number of users may mean to exhaust memory quickly or even worse swap it around. All this is remedied by caching. Diez -- http://mail.python.org/mailman/listinfo/python-list
