• Ring topology
• Two-step scheduling frameworks
• Bus topology
• network-on-chip
• wavelength assignment
• Maximum Weighted Matching (MWM)
• iSLIP
• tunable or fixed transmitter

With ever-increasing demand for high-performance computing systems, interconnection networks, serving as the communication links in multicore architectures have become a key element for guaranteeing the system performance. Compared with bandwidth-limited power hungry electrical interconnection networks, optical integrated interconnection networks also referred to as networks-on-chip (ONoC) architectures are emerging as a promising alternative to enable future computing performance.

In ONoC architectures, scheduling algorithms are necessary for avoiding packet collisions while achieving high throughput, low latency, and good fairness. Scheduling algorithms exist for non-blocking electrical NoC. These algorithms can be applied to ONoC, while accounting for additional constraints arising from optical component limitations. In this thesis various scheduling algorithms are simulated, With the objective of comparing
their latency and throughput using C + + programming language for ONoC with bus and ring topologies.

An optimal scheduler based on two-step scheduling (TSS) technique is proposed. The optimal TSS models the scheduling problem in two steps for ONoC. The first step is the matching step which is done by representing each node pair as input bipartite graph then matching takes place between the input and output ports. The second step performs the
wavelength assignment between each paired node while avoiding collisions and also with the consideration of wavelength continuity. The two-step approach with the iSLIP and MWM algorithms are considered.

The proposed optimal TSS is simulated and its performances are evaluated. The optimal scheduler with maximum weighted matching (MWM) scheduling policy achieves better results in comparison to iSLIP scheduling policy based on queue length under any packet arrival process. The optimal MWM scheduling policy achieved better performance for both
bus and ring topologies.

The main result is that unidirectional ring topology outperforms the bus topology for any
number of wavelengths less or equal to the number of ONoC port, even if the average path
length is longer. The reason is that in the bus topology half of the wavelengths are allocated in
each direction, fixing the maximum number of packets in each direction using two transceivers
per node can compensate this issue, reaching to better performance than the ring.