Method and resource manager for scheduling of instances in a data centre

A method, a resource manager, and a computer program are described for scheduling instances in a logical server, where each instance belongs to a class of instances, where each class is associated with a required service level objective (SLO). Based on determined margins between current SLOs and required SLOs, instances are sorted. If resources are not enough, an instance having a large margin may be pre-empted irrespective of the class. If available resources are sufficient for an instance having a small margin, this instance can be allocated resources. If the available resources do not suffice, it may be decided to pre-empt another instance to which resource are allocated, in order to free resources sufficient for the instance having the small margin. Accordingly, scheduling is performed for instances having various classes without violating the SLOs of the instances and such that resources are more efficiently used.