Relative Speed and Memory Allocation
 

Hi everyone,

i am trying to start a simulation with 4 Million Hexas (3 Domains) using 3 Computers (Windows XP, 2GB RAM). The computers have different speeds, so in the host.ccl three different relative speeds are set. Now when i start the solver, it seems like the solver tries to allocate memory with the same factors like in the host.ccl. The problem is that in this case not enough memory is available. Is there a way to force the solver to allocate on all computers the maximum memory, because otherwise the simulation stops with a not enough memory error.

Thanks Kramer

Re: Relative Speed and Memory Allocation
 

I think what you're basically saying is that the fastest computer is being allocated more than 2GB of memory, hence the simulation stops with a memory error. The only way to get around this is to make all the computer speeds the same in the hosts.ccl file. This will result in equal partitioning for each computer and hopefully overcome the memory error since each computer will be allocated a partition which takes less then 2GB's of memory to solve. However, you will not get good parallel speed up doing this, since and the solution will be heavily dependent on the speed of the slowest computer! How about just having a few more PC's in your cluster, with the correct speeds set in the host.ccl file, this will lower the memory requirement for each PC. You should be able to run your 4,000,000 hex problem on 4 PC's using single precision without any issues.

Re: Relative Speed and Memory Allocation
 

Hi,

Is the simulation stopping in the solver or the partitioner? If the simulation is stopping in the partitioner then simply using another partitioning algorithm will help.

Glenn Horrocks

Re: Relative Speed and Memory Allocation
 

Bank Darlehn Info Banken Geld Berechnen Schweizerkredit Konto Privatdarlehen Schufafreie Kredite in bar Kreditvermittlung