The use of optical fiber for the transmission of communication signals has been rapidly growing in importance due to its high bandwidth, low attenuation, and other distinct advantages including, radiation immunity, small size, and lightweight. Data center architectures using optical fiber are evolving to meet the global traffic demands, and the increasing number of users and applications. The rise of cloud data centers, and in particular the hyperscale cloud has produced significant changes in the enterprise IT business structure, network systems, and topologies. Moreover, cloud data center requirements are impacting technology roadmaps and standardization.
The wide adoption of server virtualization and advancements in data processing and storage technologies, have produced the growth of East-West traffic within the data center. Traditional three tier switch architectures comprising Core, Aggregation, and Access (CAA) layers, cannot provide the low and equalized latency channels required for East-Wes traffic. Moreover, since the CAA architecture utilizes spanning tree protocol to disable redundant paths and build a loop-free topology, it underutilizes the network capacity.
A new data center design called the Clos Network-based. Spine-and-Leaf architecture was developed to overcome these limitations. A Clos network is a multilevel circuit switching network, introduced by Charles Clos around 1952. Initially, this network was devised to increase the capacity of crossbar switches. It became less relevant due to the development and adoption of very large scale integration (VLSI) techniques. The use of complex optical interconnect topologies initially for high performance computing (HPC) and later for cloud data centers, makes this architecture relevant again. The Clos network topology utilizes two types of switches, Spine and Leaf Each Spine is connected to each Leaf. The work can scale horizontally minimizing the latency and non-uniformity by simply adding more Spine and Leaf switches.
This architecture has been proven to deliver high-bandwidth and low-latency (only two hops to reach destination), with the ability to provide low oversubscription connectivity. However, for large numbers of switches the architecture requires a complex mesh with large numbers of fibers and connectors, which increasing the cost and complexity of the installation, decreasing the reliability of the network.
The deployment and scaling of data center networks using transpose boxes has been disclosed in previous art. The transpose box connects switches in different layers of the mesh network. However, scaling using optical fibers does not reduce the complexity of the interconnections or improve reliability.
Future data centers will require more flexible and adaptable networks than that which can be provided by traditional mesh approaches. Highly distributed computing, high levels of virtualization, and data replication, where VMware travel seamless along the physical infrastructure. A new mesh method and apparatus that utilize spatial and spectral paths to improve connectivity in data center, is disclosed here.
Disclosed is a spatial-spectral optical mesh method and apparatus comprising an array of optical filters, optical connectors, and mirrors arranged such that positions in the vertical and horizontal dimensions are coincident, and where said optical elements satisfy specific spectral and angular and Bragg conditions. Said apparatus also comprises duplex or parallel sets of connectors and adaptors for input or output ports, and a set of collimating elements for directing light through one or more arrays of optical filters. Exemplary configurations that perform equivalent functionality are described.
The present invention, describes an apparatus and design methods to enable efficient multi-port bi-directional connections using mesh topologies required in current and future data centers networks, and HPC applications. The apparatus according to the present invention uses optical technologies, i.e,. thin films or holographic optical elements implemented in 2D or 3D configurations, are disclosed here in configurations that enable large numbers of connections with less inter-connecting fibers. Systems and methods in accordance with the present invention leverages the wide adoption of multi-wavelength transceivers utilizing wavelength division multiplexing, i.e., SWDM, CDWM, DWDM, or WDM using 4, 8 or more wavelengths.
The disclosed apparatus can implement complex network topologies inside compact small form factor modules such as Panduit's HD Flex Cassettes, enabling the efficient scaling of data centers with less fiber and connections, and higher reliability.
The apparatus according to the present invention can assist in constructing leaf-spine data center fabrics using current 100 Gb/s CWDM4, SWDM4 or future transceivers operating at ≥200 Gbps. For example, transmission speeds of 200 Gbps (2 fiber-pairs 4 wavelengths, NRZ, 25 GBaud or 1 fiber-pair, 4 wavelengths, PAM-4, 25 GBaud), 400 Gbps (4 pairs, 4 wavelengths, NRZ 25 Gbaud), 800 Gbps, (4 pairs, 4 wavelengths, PAM-4,25 GBaud), 1.6 Tbps (4 pairs, 8 wavelengths, PAM-4, 25 GBaud, or 4 pairs, 4 wavelengths, PAM-4, 50 GBaud) or 3.2 Tbps (4 pairs, 8 wavelengths, PAM-4, 50 GBaud) among many other combination of number of fibers, allocated spectrum, modulation format and symbol rate.
Examples of several embodiments of the present invention are illustrated in
The input and output ports can utilize parallel connectivity such as MPO connectors, 106, 110. In the example shown in
In this example, the first reflector, 201, reflects only λ1 and passes λ2, λ3, and λ4. The second reflector is tuned to reflect different wavelengths, λ2, λ3, and λ4 for 203, 205 and 207 respectively. Various optical filtering technologies can be utilized, i.e., thin film filters, or holographic elements. The reflective optical pass-band window matches the separation of the transmitted wavelength with a specified guard band to compensate for thermal drift, or nominal wavelength variation of the transceivers. In 209, the combined spectrum of the reflected signal is shown.
Details of array 114 are shown in
The operation of the reflector array that corresponds to MSM-4P, shown in
P
ij
=k mod(j+i−2,Nλ)+1
where, is the position of the mirror in row i and column j, k is a proportionality factor, and Nλ is the number of wavelengths, 4 in this example.
The reflectors separate and recombine wavelengths of different input ports to the associated output ports 309, 311, 313 and 315. The relationships between input and output ports can be grouped per each transmitted wavelength as shown in table I.
Also, it is noted that all the nodes launch or receive exactly, Nλ=4 wavelengths, which efficiently matches all the wavelength of the transceivers utilized. This four-port mesh is produced without fibers. The fibers are only used to connect the leaf switches to inputs of the MSM-4p and the spines switches to the output ports of the MSM-4P. It should be noted that the terms input and output ports are interchangeable here since links are bi-directional as shown in
In this figure, an array of 16 elements using 4 types of multi-wavelength reflectors, 401, 403, 405 and 407, are designed to reflect 4 wavelengths and separate and recombine up to 16 parallel ports. The interconnection map per wavelength is shown in table II.
The heights are calculate following a modular addition or subtraction as described previously for the case of the reflector position. In this design, the input signals are feed horizontally (plane XY) whereas the output signals are collected in the vertical axis. It should be noted that due to the different heights, the reflected light does not hit more than 4 reflectors. The interconnection map per wavelength is shown in table III,
By adding tunability to the reflectors using thermal, electrical (LCD), or mechanical mean, a reconfigurable circuit based switch can be produced. Even if the MSM is passive, the servers can select the required ports to direct the signal, providing the means to switch the wavelength in which the data are being transmitted.
This application claims priority to U.S. Provisional Application No. 62/696,907, filed Jul. 12, 2018, the subject matter of which is hereby incorporated by reference in its entirety.
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/US2019/041377 | 7/11/2019 | WO | 00 |
Number | Date | Country | |
---|---|---|---|
62696907 | Jul 2018 | US |