Possible enhancement: parameters related to TM1.5 performance #5
Description
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.