DPDispatcher1.1.0 actual cluster workflow teaching. run.sh runs3localCPUdimer tasks and submits3P100 +3V100GPU tasks viaSlurm, withone controlledV100firstattemptfailureandoneautomaticretry (7schedulerjobs total). It also performsactualcoldrecoveryandpartialtransferdrillusing15exactoriginalownresults;no simulationrerun for the transferdrill. Requiresallocatedaccount/resourcesandthegivencluster modules;it is not a purelyofflineplotcommand. AllscienceUID is the ordinary invokingaccount. GPUtasks1CPU1GPU;everydo_submit includingretryreadsTCmembershipfresh, excludesTCphysicalnodesandV10032. NoSSHto compute. Do not change machine/task paths during recovery. LocalContext moves outputfiles tolocal;clean=False doesnotmeanremoteoutputbackup exists. A separatelypreserved localbackup is needed torestore an accidentallydeletedcompletedoutput; this is not remote-redownload support. All generatedfiles limitedtothis extracteddirectory. Expecteddiagnosticpause onlyintransfer_drillstop; resume restores15originalSHAs withsamejobIDs. run.sh fails on existing outputdirs:newexerciseusefreshcopy. UnavailableexternalPBS/LSF/cloudservicesare not claimed tested. No sharedpackagesorserviceschanged. CPUandGPUenergies/forces use2particleLJclosedform;not a performancebenchmark.
