Job vectors of different job types arrive according to a renewal process at a service facility. The server can simultaneously serve groups of customers according to a prescribed finite set of restrictions which define the maximal number of customers of specified types for concurrent service. Service time distributions are type dependent. The control of the system is done by changing the configuration of simultaneously served types at times when a batch of customers is completed. Service of individual customers is non-preemptive. The optimization goal is to minimize the expected total amount of work in system when the system is considered under heavy traffic conditions. The considered asymptotically optimal policy is of longest-processing-time-first structure. The qualitative results compare the expected total amount of work in system for this discipline with its preemptive processor sharing counterpart.