A little more on #797

I've just pulled down beta4, but haven't had time to install it yet.  I have
had a read through the release notes, though, and was struck by the write-up
on #797 ('Uninstalling packages has become very slow').

------------------
Actually it is not significantly slower than in D4 bearing in mind that the
majority of the time is taken in tracing package pre-reqs and there are
significantly more packages (containing significantly more code) in D5 than
D4. On my machine a slightly modified version of this test took about 12
seconds to run in D4 and 21 in D5.
------------------

21 seconds ?  It takes 2 minutes 15 seconds on this machine !  I know that
you are using faster machines than this one (a  rather quick 650MZ P3), but
I wouldn't expect much more than a 2-1 difference in real speed (if that).
A factor of over 6 difference seems suspicious.

The 2:15 figure was measured with no browsers open at all (and none of my
code that observes system events installed either).   Even in beta 2, the
time taken for the same script was "only" (ha!) 2:05, but that was done
interactively, with at least the PB open, and with further sub-second delays
as I responded to each prompt.

I'll investigate further, with beta4, and let you know if I can find
anything, but in the meantime do you have any suggestions for places to look
?

    -- chris

"Chris Uppal" <[hidden email]> wrote in message
news:[hidden email]...
> I've just pulled down beta4, but haven't had time to install it yet.  I
have
> had a read through the release notes, though, and was struck by the
write-up
> on #797 ('Uninstalling packages has become very slow').
>
> ------------------
> Actually it is not significantly slower than in D4 bearing in mind that
the
> majority of the time is taken in tracing package pre-reqs and there are
> significantly more packages (containing significantly more code) in D5
than
> D4. On my machine a slightly modified version of this test took about 12
> seconds to run in D4 and 21 in D5.
> ------------------
>
> 21 seconds ?  It takes 2 minutes 15 seconds on this machine !  I know that
> you are using faster machines than this one (a  rather quick 650MZ P3),
but
> I wouldn't expect much more than a 2-1 difference in real speed (if that).
> A factor of over 6 difference seems suspicious.

Just ran the script again in a freshly installed D5 Pro beta 4 image (on a
Dual AMD MP1900+) in 20780 milliseconds, i.e. approx 21 seconds.

This was with no browsers open.

Repeating it with a package browser open on the image root, displaying the
classes pane took it up to 66985 milliseconds. Is it possible that you had a
zombie browser around somewhere?

When investigating 797 I profiled the script, and found that the majority of
the time (I forget the precise figure) was consumed by the tracing of
pre-requisites, with this happenning repeatedly, once for each package
uninstalled. As the close note says, the only really effective solution is
to calculate the net of packages that need to be uninstalled just once,
which is the subject of #508. While on that subject the reason it hasn't
been done yet, is that the algorithm for working out the package net so that
they are uninstalled in the right order is a bit tricky.

Regards

Blair

Blair,

> > 21 seconds ?  It takes 2 minutes 15 seconds on this machine !  I know
that
> > you are using faster machines than this one (a  rather quick 650MZ P3),
> but
> > I wouldn't expect much more than a 2-1 difference in real speed (if
that).
> > A factor of over 6 difference seems suspicious.
>
> Just ran the script again in a freshly installed D5 Pro beta 4 image (on a
> Dual AMD MP1900+) in 20780 milliseconds, i.e. approx 21 seconds.
>
> This was with no browsers open.
>
> Repeating it with a package browser open on the image root, displaying the
> classes pane took it up to 66985 milliseconds. Is it possible that you had
a
> zombie browser around somewhere?

That's unlikely since it was a completely fresh image.  But it's now
academic -- the same script (slightly extended to cover all the samples in
beta3/4) just ran in 50 seconds on beta4.  I still don't know what was going
wrong, but it seems to have stopped.  Which is good enough for me...

> When investigating 797 I profiled the script, and found that the majority
of
> the time (I forget the precise figure) was consumed by the tracing of
> pre-requisites, with this happenning repeatedly, once for each package
> uninstalled. As the close note says, the only really effective solution is
> to calculate the net of packages that need to be uninstalled just once,
> which is the subject of #508. While on that subject the reason it hasn't
> been done yet, is that the algorithm for working out the package net so
that
> they are uninstalled in the right order is a bit tricky.

No hurry.  I was more concerned that the timing anomaly indicated some sort
of bug.

> Blair

    -- chris