Create a datasynapse like compute grid with kubernetes


I’m looking for good resources on running jobs in a kid cluster. My specific usecase is to provide a datasynapse like compute grid for jobs that are I dependent and don’t communicate with each other. We just need fast autoscaling on both ways (scale up and shrink down). The idea is to run some computation on a cluster but we want to minimize any payments after the job is done. By the way, these are not long running processes and the processing time depends on the size of input data.
I appreciate any suggestions.