For Link21 we've been using larger skims for our CT-RAMP runs, and it would be helpful to describe various parameters that one could tweak to accommodate and/or optimize larger model runs - specifically the JPPF -Xmx parameters and the household chunk size.
Reading in larger skim files requires allocating a higher memory cap (-Xmx) to the Matrix Manager. This can be done in CTRAMP\runtime\JavaOnly_runMain.cmd in the section relating to the Matrix Manager.
For larger model runs, Node 0 also needs additional memory. The -Xmx parameters need to be adjusted accordingly in both CTRAMP\ runtime\config\jppf-node0.properties and CTRAMP\runtime\JavaOnly_runNode0.cmd. We identified some numbers that worked for us through trial and error but there might be some rules of thumb...
When running with a higher SAMPLESHARE, we changed the distributed.task.packet.size in CTRAMP\runtime\mtcTourBased.properties to speed up the model runs. (EDIT: this hasn't been as useful as we thought it would be.) We are not sure if this setting causes more memory usage at the moment.
For Link21 we've been using larger skims for our CT-RAMP runs, and it would be helpful to describe various parameters that one could tweak to accommodate and/or optimize larger model runs - specifically the JPPF -Xmx parameters and the household chunk size.
Reading in larger skim files requires allocating a higher memory cap (-Xmx) to the Matrix Manager. This can be done in
CTRAMP\runtime\JavaOnly_runMain.cmdin the section relating to the Matrix Manager.For larger model runs, Node 0 also needs additional memory. The -Xmx parameters need to be adjusted accordingly in both
CTRAMP\ runtime\config\jppf-node0.propertiesandCTRAMP\runtime\JavaOnly_runNode0.cmd. We identified some numbers that worked for us through trial and error but there might be some rules of thumb...When running with a higher
SAMPLESHARE, we changed thedistributed.task.packet.sizeinCTRAMP\runtime\mtcTourBased.propertiesto speed up the model runs. (EDIT: this hasn't been as useful as we thought it would be.) We are not sure if this setting causes more memory usage at the moment.