[Prev][Next][Index][Thread]
Re: Matt, anybody... E/L vs E/D?
>>>>> "Michelle" == Michelle Pankowski <d9060469@helios.usq.edu.au> writes:
In article <Pine.HPP.3.91.960710223755.28583C-100000@helios.usq.edu.au> Michelle Pankowski <d9060469@helios.usq.edu.au> writes:
Michelle> Here are my Executor 2 speedo test results under linux
Michelle> 2.0 and Windows95. (on a 486dx2-50 with a slooo...w hard
Michelle> drive)
Michelle> Windows95 linux-svga linux X
Michelle> video 7.6 4.6 2.7
Michelle> cpu 8.8 7.9 7.9
Michelle> disk 2.2 2.0 2.0
Michelle> My question is this;
Michelle> Linux 2.0 is linux at its finest, all 32 bit and really
Michelle> fast drivers, (tested with the fastest X server
Michelle> available too). Windows95 is all 16 bit with a .hfv
Michelle> volume to slow thing down as well...
Michelle> So why the hell is it so much faster in every damn area,
Michelle> and would it completely destroy Linux when the VCPU and
Michelle> a 32 bit port comes on line?
It's hard to say without looking carefully at your system. One thing
to bear in mind is that the Linux system might have been doing other
things at the time, since it's sometimes hard to tell what is going on
behind your back. Here are the results of me running Speedometer 3.23
under DOS mode of Windows '95 and under Linux (kernel 2.0.3) on the
same 133 MHz P5 laptop. All tests were run for 10 iterations:
Linux X DOS/Windows'95
cpu 38.444 37.608
video 7.443 19.593
disk 18.011 2.445
math 80.777 81.695
I didn't run Linux-SVGA, because currently SVGAlib doesn't know that
the Cirrus 7548 chip can be driven like a Cirrus 5428, and when it
does, that will make a big difference.
On my system, Linux did slightly better in CPU and slightly worse in
math. The video difference is the difference of being able to
directly access the screen, and the disk difference is that under
Linux, when writing to a non-HFV, we don't flush the disk cache when
new files are created, so every iteration of the disk test after the
first one just whizzes by.
Michelle> Maybe you should write a 16 bit version for Linux to see
Michelle> if you could make it go faster. (Yes, I know, it was a
Michelle> joke)
Michelle> I only tried linux to see if I could wring a few more
Michelle> bits/sec out of Executor. Tried and failed that is;)
If you run multiple iterations you might see a dramatic speedup
(depending on how much free memory you have) in the disk access. Once
we support (and the appropriate X servers support) direct video access
under X, then the video speeds will equalize.
Michelle> Makes you wonder why the linux/os2'ers are always
Michelle> bagging Windows95. It is also by far the easiest
Michelle> platform to print out PostScript files when you don't
Michelle> have a postscript printer.
We drive our non-PostScript printer using GhostScript. Granted, it
took a little fiddling to get it set up correctly, but it works fine
for us, although most of us came from a UNIX background.
I prefer Linux, but I certainly don't claim that it's the right
solution for everyone -- heck, I don't even claim that Executor is the
right solution for everyone, and I have a vested interest in Executor.
Michelle> Just the facts maam... Michelle;)
I'm still surprised that W'95 gave you a 10% gain in CPU. One thing
that could conceivably be slowing you down would be if you were short
on memory and Linux was doing various disk stuff (writing old dirty
buffers or prefetching) behind your back as you were running the
benchmark.
--Cliff
ctm@ardi.com
References: