As we have stated above, synchronous optical packet switching network is based on switching of constant length optical packets. The architecture of such network should take into account this fact. Ingress node into proposed network transforms incoming flows into optical packets with format depicted. Optical packet is composed of header and of load. Header and payload are separated by guard time fields. These fields allow us secure delimitation of the header from payload and payload from next header. The beginning of both header and load is composed of synchronization bits used for synchronization in nodes.
Header contains routing information and is 64 ns long. Guard times are 50 ns long each, yielding in 100 ns per packet. The rest of the packet is composed of data. Switching node is depicted in Figure above. Optical packets enter to the node and are firstly synchronized. Header is afterwards delimited from payload. Payload enters delay lines, where it waits until the header is processed and switching matrix is parameterized. Afterwards, payloads enter the switching matrix and are routed to the outgoing port. Before sending packets, new header is generated and concatenated to the header to the beginning of the optical packet. Optical packets are of the fixed length and therefore they are transported within fixed length time-slots. In the same time, it is impossible to guarantee that packets enter the node synchronized. Delay variation is affected by many variables, and the most important is the temperature. Therefore, there is a need for synchronizations. Synchronization can be divided into two groups: coarse and fine. Another important issue is the performance of optical switching matrix, when there are no optical memories. To improve this parameter, recirculating lines are used. Recirculating lines are intended for packets, which do not succeed to find free outgoing port. Such packets enter recirculating lines and they appear at the node entry several time-slots later.