Out of curiosity, I'm interested to understand how "dispatching" and "un-dispatching" works when WLM is capping work in a resource groups to achieve a maximum limit.
WLM starts capping work in a resource group when it is exceeding the assigned CP limit. It will setup 256 slices and decide during which slices workload will be allowed to run and during which the workload will be set non-dispatchable. Q1: At the beginning of an awake slice (i.e. when work is allowed CPU access), WLM needs to make all work units dispatchable again. Will this be done via SRB schedule into the address spaces / enclaves to turn off the non-dispatchable flag? Q2: At the beginning of a cap slice (i.e. work is no longer allowed to run), will work unit that happen to currently be queued on the work unit queue (WUQ) be actively removed as part of setting them the non-dispatchable? OR will the be allowed to sit there until they reach the top and get access to the processor? Meanwhile an SRD bill have set the non-dispatchable flag, so when the work unit gives, it will become non-dispatchable? BTW, I have read the excellent WLM paper "Resource Groups and how they work" from Laura Blodgett / Dieter Wellerdiek. But this aspect is not discussed there. -- Peter Hunkeler -- Peter Hunkeler ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
