Optical switching technologies represent a promising solution for data center interconnection networks to support the increasing bandwidth requirements of current cloud-based applications, while reducing interconnection complexity and energy consumption. Furthermore, the heterogeneity of intra- and inter-data center traffic characteristics requires some form of quality of service management. This paper describes modeling and design aspects of data center optical interconnections with particular emphasis on the aggregation level, where hybrid switching and packet scheduling are jointly applied to effectively implement service differentiation. Priority scheduling of three different service profiles is applied to maximize intra- and inter-data center traffic throughput, while guaranteeing time transparency for delay-sensitive services and zero loss/fixed delay for guaranteed connections. An analytical model is defined and validated to assess loss of real time and throughput of best effort traffic, in asynchronous packet context, when considering best effort traffic saturating the channels of the optical link. The model can also be used to dimension the optical output interface of the aggregation level switch.