SOn 22/02/21 12:14 am, tim Rowledge wrote:
> My thought these days is to skip the attempts to have a single memory
> space shared between threads/cores/whatever and have many separate
> systems running and communicating. Yes, it may be 'less performant'
> to pass bits down whatever variety of damp string required, but so
> what. Computers are better than 10 million times faster today than
> they were when we did BrouHaHa on RISC OS and the Active Book. Let's
> have a 512 core AARCH64 machine running 1500 images. In your pocket
> (admittedly right now a fairly large and well cooled pocket).

Pretty close! Many peripherals today are smart and come with their own 
cores and images. My linux desktop already has over 1400 'firmwares' 
waiting to be downloaded on demand. Smartphones, too, have tens of smart 
peripherals with multiple cores and custom firmwares.

The main problem with RAM (memory) is not that it is single space. RAM 
is 'dumb'. Having a core pull a word from RAM just to increment it is 
such a huge waste of cycles and energy. If primitive ops like 
incr/decr/complement/add etc. can be delegated to RAM (smart RAM) then 
the overall computation will become faster and energy efficient.

Linux kernel itself is small enough that RAM capacity or cost is no 
longer a constraint. Most of the bulk in a distro is in the GNU app 
stacks and device drivers.

That leaves power (battery). If power can also be smartly distributed, 
say at 8-node cluster level, then we can expect to see a cool ;-) pocket 
computer running 1500 images concurrently.

After all, Masashi Umezawa's NetMorph demonstrated seamless migration of 
live objects across closely coupled Smalltalk images way back in OOPSLA 
2002!

Regards .. Subbu