Clusters, grids and Squeak
Michael van der Gulik
squeakml at gulik.co.nz
Wed Apr 20 09:00:51 UTC 2005
Cees De Groot wrote:
>>I suspect that launching 96 images at the moment might be somewhat
>>unfeasible, as an out of the box image takes about 25 meg of memory,
>>so starting 1 per cpu would eat 2.4 Gigs before you've even started!
> Err... if you have a 96-way processor, I assume you will have a couple
> of Gigs of RAM in there....
> Of course, there are the 'small image' projects. A 2Mb image is easy,
> and Craig showed that a <100k image is possible. 2Mb would eat a
> meagre 200Mb on the 96way processor box, no problem at all I'd say.
> The *real* problem is finding algorithms to make use of 96 processors
> for browsing the web ;-)
I feel like I'm stating the obvious here, but wouldn't it be better to
have just one image running on those 96 processors?
I'd imagine one could do this by having many threads (or processes,
depending on underlying OS) using shared memory to access the image, and
clever locking so that they don't step on each other's toes.
If there isn't any shared memory but rather message-passing between CPUs
(e.g. in a cluster), then each node in the cluster would need to run its
own Squeak instance, and each would have enough memory to do so anyway.
More information about the Squeak-dev