MathGroup Archive 2008

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Parallelising NDSolve

  • To: mathgroup at smc.vnet.net
  • Subject: [mg90365] Re: Parallelising NDSolve
  • From: Jens-Peer Kuska <kuska at informatik.uni-leipzig.de>
  • Date: Mon, 7 Jul 2008 05:07:58 -0400 (EDT)
  • References: <g4q9pp$eb9$1@smc.vnet.net>

Hi,

the operating system determines how large the address space is.
And the operating system determines how it distribute the 
processses/threads onto the processors.

You can have a hexadeca-core processor as long as you run an 32bit
operating system you will have less than 4 Gbyte address space.

And yes, you can parallelize a initial value solver, this will need
atleast *more* memory than a serial initial value solver and
you need a very huge system (>1000) of odes to see that the parallel 
execution will be faster.

will be

Regards
   Jens

Alex Butler schrieb:
> Hi All,
> 
> Occasionally the data that NDSolve generates is too large for the 
> Windows Process Address Space of my PC. I have a dual core machine and 
> was wondering if it were possible to split NDSolve between these two 
> processors, thus doubling the amount of Process Address Space available 
> to NDSolve.
> 
> Firstly, can NDSolve be parallelised? And secondly, what's the best way 
> to go about it?
> 
> Thanks,
> Alex
> 


  • Prev by Date: Re: distribution fitting
  • Next by Date: Re: Set::setraw error
  • Previous by thread: Parallelising NDSolve
  • Next by thread: Re: Parallelising NDSolve