Slowness [RE: [Example of my Gnus Performace] gargle gargle howle howle]
ben at 666.com
Sat Nov 13 20:42:54 EST 2004
> 34 seconds really is not acceptable for that task. It could
> take as little time as it takes to read a 8000x80 (ie, 640KB)
> file and narrow the buffer to the last 2000, then fontify the
> displayed area. Ie,
> (much) less than 1 second.
> I understand why Gnus makes the tradeoffs it does, but I'm
> sure Outlook makes different ones that bring it in line with
> the 1 second estimate, and I definitely have situations (that
> occur daily) where that 30 second difference means I work
> suboptimally (or drink excessive amounts of coffee).
> We really need to do something about these speed problems if
> we want XEmacs to be acceptable as an MUA environment for a
> lot of people.
On my 700Mhz PC, if I visit my Sent Items folder [6224 messages, 165KB]
under Outlook, it takes about 1-2 seconds the first time, and then
negligible afterwards. To resort based on a different field takes another
1-2 seconds the first time, then maybe 0.3 seconds afterwards.
A lot of packages under XEmacs are just dog-slow. Visiting a Java file, for
example, takes a minute or more the first time, when all sorts of JDE stuff
loads. What I don't know is where the slowness is coming from -- e.g. if
you compare a release XEmacs to a release GNU Emacs, what's the difference?
And where is the slowness when you profile it? Almost none of these package
makers seem to bother profiling very much, even though it's very easy under
XEmacs. Sometimes when I get really annoyed I try to profile.
Unfortunately I don't really have good tools for it ...
Much of the slowness I now see shows up as "redisplay". I have a strong
suspicion that Andy's widget code is a good chunk; when I turn off the tabs,
things speed up dramatically, for example.
I would really like to see more people looking into this.
More information about the XEmacs-Beta