Datacenter and high-performance computing systems demand for interconnections between computing hosts with low latency, low-power consumption and traffic-aware adaptability. Researchers have demonstrated in recent years that all-optical networks can offer these characteristics by leveraging technologies such as microring resonators, Mach-Zehnder switches and wavelength routers interconnecting top-of-rack electronic packet switches (EPSs). One of the challenges in optical switching is the lack of optical buffers, which makes it difficult to perform optical reconfiguration without significant packet loss.In this thesis, we study the benefit of using a reconfigurable optical network with a make before break (MBB) approach to improve network performance and reduce packet loss when we perform a reconfiguration operation. We compare a plain optical reconfiguration (OST) against a MBB approach in terms of packet loss, latency and throughput as we attempt to achieve hitless reconfiguration. Understanding the transport protocol (TCP) is relevant in our experiments because the path reconfiguration triggers congestion control timers that pause traffic between hosts. Thus, a general analysis of the retransmission timeout (RTO) is presented when studying our performance metrics. Our testbed combines the advantages of electronic and optical networks: buffers, SDN integration, flexible topology and optical links. Our design is modular, hence more servers or switches can be added on demand and configured with software for faster deployments in future experiments. On average, when we tested a single stream of data between two servers, the packet loss decreased from 2.8% in OST to 0.93% in MBB. The link unavailability due to RTO events went from 598 ms down to 121 ms (80% reduction). With MBB, the throughput also improved as it dropped only by 0.6 Gbps instead of a 6Gbps hit in the case of OST.