[squeak-dev] Prepare for Thousands of Cores --- oh my Chip - it's full of cores!

Andres Valloud AVALLOUD at roadrunner.com
Mon Jul 7 23:18:32 UTC 2008


Hello,

I have not thought this through, but it does make me curious.  If I 
understand this right, queue theory says that the time it takes to 
service an item in the queue is proportional to the number of things in 
the queue.  It seems to me that jobs one might want to throw at a native 
thread also go into a queue of things waiting to be served by the CPU.  
How does queue theory play out when a CPU has 1000 cores, I presume at 
the very best 2x the memory bandwidth we have now, and most programs are 
written so that they split their work into N pieces as small as possible 
to get the most out of the available cores?

Has anybody looked into this?  Perhaps somebody read something worth 
sharing?

Andres.



More information about the Squeak-dev mailing list