Re: [Pharo-project] on Raspberry Pi
On May 10, 2012, at 1:34 PM, H. Hirzel wrote:
On 5/10/12, Pavel Krivanek <pavel.krivanek@gmail.com> wrote:
Hi,
one my friend got Raspberry Pi yesterday so I have seen it in action. I had almost no time for experiments but I have seen Fedora 17 on it with standard Squeak (3.10) packages for this distro on FullHD display. On that VM (3.10.5) I got about 40M bytecodes/sec and 1M sends/sec. The environment is not extra fast but usable without problems. I will pray for successful GSoC ARM JIT project :-) The device was connected to the professional meter and the power consumption was about 3.5W.
Cheers, -- Pavel
Amazing, a 256MB RAM machine with a main board which includes everything for 35USD. Some five to seven years ago one would not have believed that this is possible.
This especially *screams* for re-considering certain design decisions that we inherited from Smalltalk. 256MB of RAM makes it hard to argue for complex schmemes to save some a little ram. (Compact classes, for example). How would smalltalk look like if it would be re-designed with the same philosophy that they used when there where just 256Kb in the Alto? This is a fun comparision: http://www.raspberrypi.org/ vs http://en.wikipedia.org/wiki/Xerox_Alto 1) Price. $35 vs. $35000 (est, not adjusted for inflation). x1000 2) RAM 256MB vs. 256Kb (max was 512, at the end). x1000 3) CPU 1MIPS? not sure. Vs. 700Mhz ARMv11 x 1000 (?) + 3D and MP4 decoder So one could argue that it's a factor of 1 Million considering price and Performance together... Marcus -- Marcus Denker -- http://marcusdenker.de
Yes this is an interesting comparison, in particular if you add size and weight to the comparison. Thank you Marcus, for the idea of doing the comparison putting the figures together. And yes, 1000 times more RAM should make it possible to come up with a more easy to understand design for the object memory and VM. --Hannes On 5/10/12, Marcus Denker <marcus.denker@inria.fr> wrote:
On May 10, 2012, at 1:34 PM, H. Hirzel wrote:
On 5/10/12, Pavel Krivanek <pavel.krivanek@gmail.com> wrote:
Hi,
one my friend got Raspberry Pi yesterday so I have seen it in action. I had almost no time for experiments but I have seen Fedora 17 on it with standard Squeak (3.10) packages for this distro on FullHD display. On that VM (3.10.5) I got about 40M bytecodes/sec and 1M sends/sec. The environment is not extra fast but usable without problems. I will pray for successful GSoC ARM JIT project :-) The device was connected to the professional meter and the power consumption was about 3.5W.
Cheers, -- Pavel
Amazing, a 256MB RAM machine with a main board which includes everything for 35USD. Some five to seven years ago one would not have believed that this is possible.
This especially *screams* for re-considering certain design decisions that we inherited from Smalltalk.
256MB of RAM makes it hard to argue for complex schmemes to save some a little ram. (Compact classes, for example).
How would smalltalk look like if it would be re-designed with the same philosophy that they used when there where just 256Kb in the Alto?
This is a fun comparision:
http://www.raspberrypi.org/ vs http://en.wikipedia.org/wiki/Xerox_Alto
1) Price.
$35 vs. $35000 (est, not adjusted for inflation). x1000
2) RAM
256MB vs. 256Kb (max was 512, at the end). x1000
3) CPU
1MIPS? not sure. Vs. 700Mhz ARMv11 x 1000 (?) + 3D and MP4 decoder
So one could argue that it's a factor of 1 Million considering price and Performance together...
Marcus
-- Marcus Denker -- http://marcusdenker.de
On 10 May 2012, at 14:03, H. Hirzel wrote:
Yes this is an interesting comparison, in particular if you add size and weight to the comparison.
Thank you Marcus, for the idea of doing the comparison putting the figures together.
And yes, 1000 times more RAM should make it possible to come up with a more easy to understand design for the object memory and VM.
Have you seen SOM (Simple object machine)? That's what you get if you go for simplicity for teaching. See http://www.hpi.uni-potsdam.de/hirschfeld/projects/som/, it is originally coming from à rhus. It is a nice and clean VM, with little distracting complexity. Not fast of course... And, well, not Pharo compatible. (which one could count as a benefit, since it keeps complexity low) Would be interesting to hear what the Pinocchio folks think about it in terms of complexity. Best regards Stefan -- Stefan Marr Software Languages Lab Vrije Universiteit Brussel Pleinlaan 2 / B-1050 Brussels / Belgium http://soft.vub.ac.be/~smarr Phone: +32 2 629 2974 Fax: +32 2 629 3525
Thank you, Stefan for this link giving a list with short explanations of different compatible Smalltalk VMs. I was not aware of it. Especially interesting is the diagram in the history section of http://www.hpi.uni-potsdam.de/hirschfeld/projects/som/ The VM called 'AweSome' implemented in Squeak (includes C++ generator) has 2432 LOC whereas the Java implementation has 5899 LOC. --Hannes On 5/10/12, Stefan Marr <smalltalk@stefan-marr.de> wrote:
On 10 May 2012, at 14:03, H. Hirzel wrote:
Yes this is an interesting comparison, in particular if you add size and weight to the comparison.
Thank you Marcus, for the idea of doing the comparison putting the figures together.
And yes, 1000 times more RAM should make it possible to come up with a more easy to understand design for the object memory and VM.
Have you seen SOM (Simple object machine)?
That's what you get if you go for simplicity for teaching.
See http://www.hpi.uni-potsdam.de/hirschfeld/projects/som/, it is originally coming from à rhus.
It is a nice and clean VM, with little distracting complexity. Not fast of course... And, well, not Pharo compatible. (which one could count as a benefit, since it keeps complexity low)
Would be interesting to hear what the Pinocchio folks think about it in terms of complexity.
Best regards Stefan
-- Stefan Marr Software Languages Lab Vrije Universiteit Brussel Pleinlaan 2 / B-1050 Brussels / Belgium http://soft.vub.ac.be/~smarr Phone: +32 2 629 2974 Fax: +32 2 629 3525
participants (3)
-
H. Hirzel -
Marcus Denker -
Stefan Marr