I'm pretty sure the RAM in the WinCE machines is
some kind of RAM--probably not ordinary DRAM,
for power reasons, but not Flash RAM either. For
one thing, you lose your data when the battery dies.
For another, I think you need a lot of RAM to run
even the stripped-down Micros applications that come
with WinCE.

However, I'm sure you are correct in suspecting a
bottleneck between Squeak and memory that slows
performance. It could be that the memory bus is
only 16- or even 8-bits wide. The bus is probably quite
slow and may need to be shared with the display refresh
process, so Squeak can only get a fraction of the full
bus bandwidth. Finally, these machines don't have L2
caches (far too much of a power drain) and they may
have insufficient on-chip caches to support Squeak.
(Does anyone know what size caches the Hitachi SH3
or MIPS 4000 chips have?)

Thanks for benchmarking '26 benchfib' on the Casio E-10.
If anyone else has Squeak running on a WinCE machine,
I'd love to know how long the following benchmarks take:

	26 benchfib
	10 benchmark

These methods are retained even in the TinySqueak image.
The first measures message send speed, the second is a
good predictor of general bytecode execution speed.

Thanks!

	-- John
 
ohshima at is.titech.ac.jp wrote:
>   I've been wondering what kind of memory they are which is
> called "RAM" on the CE machines.
> 
>   They can store data (of course), and the data are kept
> while the machine is turned off.  The user can divide the
> into a part for data and another part for program execution.
> And especially if the memory space is tight, the execution
> speed of the applications become so sloooow.
> 
>   When I took a look at Squeak 2.2 for WinCE on Casio E-11, 'WarpBlt
> test3' ran pretty fast, but the performance of benchFib was
> quite bad. (about 60 seconds for '26 benchFib'.)
> 
>   I suspected that something was wrong with the access speed
> for data, but I couldn't figure out what is happening.
> ...