Re: Speed of linux

[John Logsdon <j logsdon lancaster ac uk>]

On Tue, 2 Feb 1999, John Wilkinson wrote:

> Hi all,
> I did ask a similar question the other day. It is sad that Linux is not
> at least as fast as W95.

This is a big puzzle and I suppose most people have concluded that there
is something funny with your setup but haven't been so rude as to suggest
it to you!  In my experience, Linux is much more efficient in memory
handling, scheduling and virtually everything else than anything Microsoft
offers.  W95 in particular is supposed to be very slow cf NT or 98 - not
much better than W3.11 for the same application, although the GUI is much
better. 

This is almost certainly not your fault!  Linux is much more testing of
the hardware that W95 and I suggest that your video card or motherboard
or some other part of your hardware are just not compatible or there is a
fault somewhere.  The fact that Scilab (which is very fast, although I
generally prefer Octave as it is more Matlab compatible) runs faster on
W95 than Linux suggests that it is something in the hardware to me.

There is another possible reason that I can think of.  A minute item on
the R list recently arose where a non-linear solution with numerical
derivatives was *wrongly* calculated under W95 but correct under Linux. 
The c and fortran sources for the two versions were identical.  This
(obviously) freak result was thought to be because of linking with some
mathematical MS DLL's which actually have an error!  So it *may* be that
the Scilab results (if they are the same) have been arrived at by some
accidental approach which leads to a different route.  Compare the results
if you can with something run on, say, a Sun. 

> If we are all so certain that it is the X windows stuff that is the
> cause, this implies that the text screen in Linux is much much faster
> than DOS. I would be interested to see some comparisons if any one has
> done this in the past.

Have you tried running Scilab from the console (ie no X) or is your
application graphical?  You can see if it is an X problem by comparing the
two - ensure that Scilab is well and truly loaded first thought by doing
the calculation twice in both console and X and ignoring the first result. 

> 
> Also,  do we know if products like accelerated X or Metro are any faster
> than XFree86? I run metro but this seems just as fast as XFree.
> 

I don't think Metro-X will be any faster than XFree but Accelerated X will
be for real graphic intensive operations like 3D.  I have seen some
stunning demos with Accelerated X (demos are always stunning of course -
you never see the failures) but I suggest that the solution is either in
you hardware compatability/fault or that you have a curious problem that
has benefited from the MS DLL problem I have heard of. 

The great diversity of hardware on which Linux runs and the general
unhelpfulness and sloppiness of hardware manufacturers is to blame.  I
always think that running such a complex operating system on a computer
that costs only a few hundred pounds/dollars/euros is just amazing and the
stability of the systems alone is worth all the headache of being a
home-grown Sysadmin! 

Good luck - the fact that you haven't given up is clear from your frequent
participation on this list!  Keep us posted if you find the solution. 

\John