Hi All,

I wrote: 


 > --- G3 333Mhz PB now gives 5.8 ticks per launch.
 > --- FX now gives 8.5 ticks per launch.

tedd then got:

 --- 6.0 ticks per launch with G4 500 Mhz

(But which system??)

I mused:

 > That [68K vs G3/4] is amazingly close and suggests 
 > that my use of a disk file to store the record
 > of launches is predominating.

In other words one way or another, we may just be testing 
hard disks.

Well at this point I believe I have gotten rid of the
effect of recording time lapse on disk It is not a big one.
Here is the simple trick:  Instead of recording in RAM as I
first suggested, I have set up a necklace of 10 programs
MultiLaunch_mini1,...,MultiLaunch_mini9,MultiLaunch_miniX
and have only the one in ten do the recording, while the
others just launch the next in the necklace. Why change
concept? Because I had essentially no coding to add and the
test is a throwaway item.  The necklace test is posted as:

"NecklaceLaunch_mini.cpt.gz.hqx" in directory:

ftp://topo.math.u-psud.fr/pub/lcs/fb/DragAndDrop/

The results bring FX and G3 still closer together.

  --- G3 333Mhz PB now gives 6.6 ticks per launch.
  --- FX now gives 7.0 ticks per launch.

I note that the difference between FX 40 Mhz and G3
333Mhz is now less than the difference engendered by
side issues like applications up, change of disk, or
extensions. (I used the same IBM 9 Gig hard disk with
HFS for both).

Better speeds in racy circumstances: 

 --- 5.1 ticks per launch,  FX 40 Mhz with *no extensions*
 --- 3.9 ticks per launch,  FX 40 Mhz 
     with *no extensions* and necklace of programs on
     a Shrinkwrap RAM disk placed in RAM

The G3 333Mhz PB cannot keep up  --- maybe because my Sonnet
accellerator card needs some extensions to function
correctly.(??)

 --- 5.8 ticks per launch, PB G3 333 Mhz with *no extensions*
 --- 4.8 ticks per launch, PB G3 333 Mhz and ShrinkWrap RAM disk
     and *a few extensions / control pannels*.

tedd > you should have the user shut down all applications, 
 > instruct the user not to do anything and even hide the cursor 
 > -- everything counts in cycles.

Amen.  The Mac is clearly not designed for racing.  Tedd should
beat the FX by a factor of > 10.

Other reactions?  

            Cheers

                   Larry



PS.  The slowest time reported so far is (as predicted)
for early PPC (PowerMac 8100 OS8.6) but it should be
rechecked with the sharper tests.

PPS. Larry's treadmill law:  Every increase in processor speed
will be matched by an equal reduction in system speed.