PRESENTACIÓN DE LAS HIPÓTESISY PROPUESTA DEL

fibres. The packet is dropped if the switch is not configured within this time. The processing speed of the switch controller should be compatible with the data rate of the incoming channel. Low processing speed of the switch controller also results in packet losses. The limitations of OPS are the lack of feasible optical buffers and packet losses in the case of output port contention or low processing speed of the switch controller. Fast optical switches are required to use OPS because of their low switching time.

The OBS[38–42] is different from OCS and OPS and is considered as a compromise

between them. It has a separate control and data plane similar to OCS. Packets are aggregated into a burst. A control packet is then transmitted on a dedicated control channel to reserve resources on all intermediate nodes from the source to the desti- nation. The burst is sent at a particular time after sending the control packet which is called the offset time. During the offset time, the burst is temporarily stored at the edge node before transmission. During this time, the switch controller at the core node processes the control information and sets up the switching matrix for the incoming burst. Burst loss due to output port contention is the major limitation of the OBS net- work. Output port contention can occur due to unavailability of a wavelength at the desired output port for an incoming burst. Several techniques exist in the literature

to avoid contention such as FDLs [43, 44], Deflection Routing [45, 46], Wavelength

Conversion [47] and Segmentation Based Dropping [48, 49] but none of them can

guarantee zero burst loss. OBS with a two-way reservation scheme ensures zero burst loss in which a control packet reserves resources across all intermediate nodes from the source to the destination and is sent back to the source as an acknowledgement. However, the control packet has a high Round trip time (RTT) for a large wide area optical network. This is because of the high propagation and switching delay.

1.5

Overview of Proposed Solution

In this thesis, novel optical interconnection schemes which are based on OBS are pro- posed. Historically, the OBS was proposed for the backbone optical core network but

1.5. OVERVIEW OF PROPOSED SOLUTION

it has not replaced OCS due to its limitation of high burst loss in this application. The proposed schemes consider OBS with a two-way reservation protocol that ensures zero burst loss. In the two-way reservation protocol, the connection is established for each burst before transmission. The two-way reservation is not suitable for long-haul backbone optical networks due to the high RTT of the control packet but for the pro- posed optical interconnect for the DCN, this RTT is not high for several reasons:1) the propagation delay is negligible, 2) faster optical switches are used at the core, 3) a fast optical control plane is used, 4) processing of the control packet is rapid and 5) a single hop topology is used. Network-level simulation is used to evaluate the performance of the proposed schemes.

In the first scheme, an optical interconnection scheme which is based on both fast and slow optical switches is proposed which is called hybrid optical switch architecture (HOSA). The proposed technique leverages strengths of both types of optical switches. The strengths of one type of optical switch compensate for the weaknesses of the other type. HOSA employs a single stage core topology that can be easily scaled up (in ca- pacity) and scaled out (in the number of racks) without requiring major re-cabling and network reconfiguration. HOSA features separate data and control planes. The control plane comprises a centralized controller while the data plane contains an ar- ray of fast and slow optical switches. The main idea is to route high volume traffic through the fast optical switch during the reconfiguration of the slow optical switch. The traffic is moved to the slow MEMS switch once it is reconfigured. This results in the overall switching time being that of the fast optical switch. A resource assignment algorithm is presented that allocates resources efficiently to ensure minimum latency. The scalability of the proposed design is evaluated by considering various capacities of servers in a rack and various ratios of fast and slow optical switches. A trade-off be- tween cost and power consumption of the proposed design using analytical modelling to compare it with conventional interconnects is also presented. Furthermore, the trade-off between the performance and the capacity of both types of optical switches is evaluated. In this scheme, large burst aggregation time is the major limitation i.e. the aggregated traffic should be high enough so that it can bypass the MEMS switch during its reconfiguration time.

1.5. OVERVIEW OF PROPOSED SOLUTION

In the second scheme, HOSA supplemented with a Traffic Demand Scheduling scheme is proposed to overcome the limitation of high aggregation time in HOSA. In HOSA with TDS, there is no need to aggregate a large amount of traffic. The con- troller maintains a traffic demand matrix which updates traffic demand on periodic intervals and assigns slow paths for the high traffic volume. A resource allocation al- gorithm is proposed in the control plane for efficient utilization of the resources that results in high throughput and low latency. However, similar to other hybrid tech-

niques[23, 50–54], HOSA with TDS has the limitation of the control plane overhead

to measure the communication patterns and calculate a new schedule. As a result, the control plane is limited to support only traffic with high stability. Stability refers to the duration of the traffic flows i.e. high stability refers to the traffic flows that last several seconds. Due to the limitation of the control plane, HOSA with TDS does not perform well for dynamically changing communication patterns. To overcome this limitation, this thesis presents a new optical interconnect called fast optical switch architecture (FOSA).

FOSA is based on only fast optical switches. In this design, only OBS with the two-way reservation is used. In FOSA, there is no extra processing overhead on the control plane for maintaining traffic demands as was done in the second scheme. This scheme works well for dynamically changing communication patterns. The proposed technique shows considerable improvement in terms of throughput and packet loss ratio compared to the traditional methods of OBS while delays comparable to those of the traditional methods of OBS are also achieved. The proposed technique also demonstrates performance comparable to that of electrical data centre networks.

Another reason that OBS has not replaced OCS in the long-haul optical network is because of its poor performance with TCP. The performance of TCP over a traditional OBS network is degraded by the wrong interpretation of congestion in the network. The burst losses due to contention can be misinterpreted by the burst losses due to con- gestion. The contention refers to the burst loss due to unavailability of a wavelength even at the low traffic load in the network. The performance of the TCP is evalu- ated on FOSA. As in the proposed scheme, the burst loss is zero due to the two-way reservation protocol; the results show significant improvement of TCP performance in