A bit more than three years ago, I renewed my main machine and bought an Athlon XP 2600+ with 512MB of RAM and a 80GB hard disk. The speed boost I noticed in games, builds and the overall system usage was incredible — I was coming from a Pentium II 233 with 384MB of RAM.
With the change, I was finally able to switch from plain window managers to desktop environments (alternating KDE and GNOME from time to time) and still keep a usable machine. I was also able to play the games of that era at high resolutions. And, what benefited me more, the build times of packages and NetBSD itself were cut by more than a half. For example, it previously needed between 6 to 7 hours to do a full NetBSD release build and, after the switch, it barely took 2. On the pkgsrc side, building some packages was almost instantaneous because the machine processed both the infrastructure and the source builds like crazy.
But time passes and nowadays the machine feels extremely sluggish. And you know that hardware does not degrade like this so it's easy to conclude it's software's fault. (Thank God I've done some upgrades on the hardware, like doubling the memory, replacing the video card and adding a faster hard disk.)
I'm currently running Kubuntu 6.10 and KDE is desperately slow in some situations; of course GNOME has its critical scenarios too. (Well... it is not that slow, but responsiveness is, and that makes a big amount of the final experience.) The problem is they behaved much better in the past yet I, as a desktop user, haven't noticed any great usability improvement that is worth such speed differences. As a side note: I know the developers of both projects try their best to optimize the code — kudos to them! — but this is how I see it in my machine.
Another data point, this time more objective than the previous one. Remember I mentioned NetBSD took less than 2 hours to build? Guess what. It now takes 5 to 6 hours to build a full release; it's as if I went back in time 3 years! Or take pkgsrc: the infrastructure is now very, very slow; in some packages, it takes more time than the program's build itself.
I could continue this rant but... it'd drive nowhere. Please do not take it as something against NetBSD, pkgsrc and KDE in particular. I've taken these three projects to illustrate the issue because they are the ones I can compare to the software I used when I bought the machine. I'm sure all other software suffers from slowdowns.
Anyway, three years seem to be too much for a machine. Sometimes I think developers should be banned fast machines because, usually, they are the ones with the fastest machines. This makes them not notice the slowdowns as much as end users do. Kind of joking.
here are a couple of noise-points which tend to the same idea.
ReplyDeleteplan9 has taken a very VERY reductionist position on #include. Their C compiler runs faster than any on the planet. Yes, optimizing compile-TIME compared to run-TIME performance seems counter intuitive, unless you are a developer: you WANT it fast to do the code-compile-test cycle over and over again.
secondly, I just timed firefox loading a file:///path/to/my.html type instance. This should be blindingly fast: there is no network delay, no DNS delay, and the file is a very simple table. Yet, it takes seconds. Not even the screen refresh should be visible, but I can watch it "think" how to do this. Bad.
It's funny that you mention build times... because I've been annoyed by them just yesterday.
ReplyDeleteI am currently adding a new feature to Boost's Quickbook utility and touching one single file makes a compile/test cycle of TEN minutes! That's unacceptable. (Well, strictly speaking this is because of a known bug in binutils' 2.17 ld which makes linking of objects built with -g extremely slow. But anyway, building without -g still keeps the whole time up to 3 minutes; rather inconvenient.)
G++ is very slow, unfortunately. Building under Windows the same code with Visual C++ or Borland C++ is noticeably faster.
And another thing I noticed just by chance. I put to build Bochs from pkgsrc under my iBook G4 (1.33Ghz and 1GB of RAM) and my testing machine (Pentium III 733Mhz and 256 of RAM) at the same time, and they both finished at the same moment. Need to do some more tests, but one'd assume that the laptop ought to be faster... if only by a small amount...