Arrayed waveguide grating based modular interconnection networks and methods for constructing and applying the same

Abstract

An arrayed waveguide grating (AWG) based interconnection network and modular construction method, comprising N1 left nodes, with each left node having N2 ports, N2 right nodes, with each right node having N1 ports, where N1≧N2, N1 and N2 having a greatest common divisor r, and each port having an optical transceiver associated with a fixed wavelength; N1n2 r×1 wavelength multiplexers having their input ports respectively connected with the ports of N1 left nodes, where n2=N2/r; N2n1 1×r wavelength demultiplexers having their output ports respectively connected with the ports of N2 right nodes, where n1=N1/r; n1n2 r×r AWGs connecting the r×1 wavelength multiplexers and the 1×r wavelength demultiplexers r×rn1n2, and each of the r×r AWGs being associated with a wavelength subset {λk|k=0, 1, . . . , r−1}.

Description

CROSS-REFERENCE AND RELATED APPLICATIONS

The subject application claims priority on Chinese patent application No. CN 201410410122.8 filed on Aug. 20, 2014. The contents and subject matter of the Chinese priority application is incorporated herein by reference.

FIELD OF TECHNOLOGY

The present invention relates to large-scale networking of a data center, and in particular, relates to an arrayed waveguide grating (AWG) based modular interconnection network and methods for constructing and applying the same.

BACKGROUND OF THE INVENTION

In the past decade, fast global information development push data centers to evolve along the direction of super large scale. In 2012, each data center of Amazon has about 60,000 servers, and that for Google, the number exceeds 50,000 servers, while Microsoft is building a data center containing over 300,000 servers. Current data centers generally adopt the layered tree topology, such as the ‘Fat Tree’ topology. The advantage of the tree topology is its large bi-directional bandwidth, while the disadvantage is its lack of scalability. As pointed out in the Article entitled “60 GHz Data-Center Networking: Wireless Worry less,” topological and cabling complexity in data centers is reaching unimaginable proportions, leading to maintenance challenges, inefficient cooling, and substantial operational costs.

To tackle the interconnecting issue, the following proposals have been put forth in the industry:

The first is to replace wires with wireless interconnects. Wireless networking has signals prone to interference, small bandwidth, and high consumption of stations, making it only a topic at the academies, and not fit for practical application.

The second is to develop a unified structure based on integrated and enhanced Ethernets, with Cisco and Brocade being the major proponents. Although the method decreases the number of cables, trunk adapters, and network interfaces, the cost of equipping and integrating network adapters is huge, further, main boards for Ethernet optical fiber channels are not yet available.

The third is to adopt structured cabling. The main idea is to divide the cabling system in a data center into a main cabling area and a device area based on different devices. The main cabling area and the device area are connected with optical cables, while servers, switchers, and storage devices are jumper-connected via cable distribution frames in the main cabling area and the device area. The proposed solution simplifies cable management to some extent, making it possible to move or modify the system by merely changing jumpers. But it does not decrease the number of system cables, thus interconnecting complexity remains and the operating difficulty is still high.

Because of the above reasons, a method of module networking based on Arrayed Waveguide Grating (AWG) and directed at cabling complexity and management in a data center is in need.

SUMMARY OF THE INVENTION

The present invention solves the problem of the cabling complexity and related management issues by providing a design method of AWG-based modular interconnection networks to reduce cabling complexity so as to simplify networking maintenance and management.

The present invention provides an AWG-based modular interconnection network comprising left nodes, the number of the left nodes being N₁, with each left node having N₂ ports; right nodes, the number of the right nodes being N₂, with each right node having N₁ ports; N₁ and N₂ each representing a positive integer, N₁≧N₂, and N₁ and N₂ having a greatest common divisor r, and each port having an optical transceiver associated with a fixed wavelength, characterized in that it further comprises:

N₂×1 wavelength multiplexers, the number of the N₂×1 wavelength multiplexers being N₁, with each of the N₂×1 wavelength multiplexers having N₂ input ports being respectively connected with the N₂ ports of each of the left nodes;

1×N₁ wavelength demultiplexers, the number of the 1×N₁ wavelength demultiplexers being N₂, with each of the 1×N₁ wavelength demultiplexers having N₁ output ports respectively connected with the N₁ ports of each of the right nodes; and

an N₁×N₂ AWG connecting the N₂×1 wavelength multiplexers and the 1×N₁ wavelength demultiplexers, the N₁×N₂ AWG having N₁ input ports and N₂ output ports, and being associated with a wavelength set Λ={λ₀, λ₁, . . . , λ_N1−1}.

The present invention further provides an AWG-based modular interconnection network, comprising left nodes, the number of the left nodes being N₁, with each of the left nodes having N₂ ports, right nodes, the number of the right nodes being N₂, with each of the right nodes having N₁ ports; N₁ and N₂ are integers, N₁≧N₂, N₁ and N₂ having the greatest common divisor r, and each port having an optical transceiver associated with a fixed wavelength, characterized in that it further comprises:

r×1 wavelength multiplexers, the number of the r×1 wavelength multiplexers being N₁n₂ (N₁n₂ stands for the product of N₁ and n₂, and same goes for the same styled numbers in the subject application), and n₂ r×1 wavelength multiplexers having their input ports respectively connected with the N₂ ports of one of the left nodes, where n₂=N₂/r;

1×r wavelength demultiplexers, the number of the 1×r wavelength demultiplexers being N₂n₁, and n₁ 1×r wavelength demultiplexers having output ports respectively connected with the N₁ ports of one of the right nodes, where n₁=N₁/r; and

r×r AWGs connecting the r×1 wavelength multiplexers and the 1×r wavelength demultiplexers, the number of the r×r AWGs being n₁n₂, and each of the r×r AWGs being associated with a wavelength subset {λ_k|k=0, 1, . . . , r−1}.

The present invention also provides an AWG-based modular interconnection network, comprising left nodes, the number of the left nodes being N₁, with each of the left nodes having N₂ ports; right nodes, the number of the right nodes being N₂ with each of the right node having N₁ ports; N₁, N₂, and K are integers, where N₁=KN₂; and each port having an optical transceiver associated with a fixed wavelength, characterized in that it further comprises:

N₂×1 wavelength multiplexers, the number of the N₂×1 wavelength multiplexers being N₁, each of the N₂×1 wavelength multiplexers having N₂ input ports respectively connected with the N₂ ports of each of the left nodes;

1×N₂ wavelength demultiplexers, the number of the 1×N₂ wavelength demultiplexers being KN₂, the K 1×N₂ wavelength demultiplexers having their output ports respectively connected with the N₁ ports of one of the right nodes; and

N₂×N₂ AWGs connecting the N₂×1 wavelength multiplexers and the 1×N₂ wavelength demultiplexers, the number of the N₂×N₂ AWGs being K, each of the N₂×N₂ AWGs being associated with a wavelength subset {λ_k|k=0, 1, . . . , N₂−1}.

The present invention further provides a method for constructing an AWG-based modular interconnection network having N₁ left nodes, with each left node having N₂ ports; the N₂ right nodes, with each right node having N₁ ports; N₁≧N₂, N₁ and N₂ being integers having a greatest common divisor r; and each port having an optical transceiver associated with a fixed wavelength, characterized in that the method comprises the following steps:

Step 1: providing N₁ N₂×1 wavelength multiplexers, labeled by L₀, L₁, . . . , L_N1−1, for the N₁ left nodes, the ith N₂×1 wavelength multiplexer having its jth input port connected with the jth port of the ith left node, and the jth port of the ith left node is associated with wavelength

${\lambda }_{{\left[i+j\right]}_{N_1}},$ where i=0, 1, . . . , N₁−1, j=0, 1, . . . , N₂−1;

Step 2: providing N₂ 1×N₁ wavelength demultiplexers labeled by R₀, R₁, . . . , R_N2−1, for N₂ right nodes, the jth 1×N₁ wavelength demultiplexer having its ith output port connected separately with the ith port of the jth right node, and the ith port of the jth right node is associated with wavelength

${\lambda }_{{\left[i+j\right]}_{N_1}};$

Step 3: interconnecting the N₁ wavelength multiplexers on the left with the N₂ wavelength demultiplexers on the right via an N₁×N₂ AWG, the N₁×N₂ AWG having N₁ input ports and N₂ output ports, and being associated with a wavelength set Λ={λ₀, λ₁, . . . , λ_N1−1}.

The method further comprises:

Step 4: substituting the N₁×N₂ AWG with a three-stage AWG network S_A(n₁,r₁,m_A,n₂,r₂), the AWG network S_A comprising N₁=r₁n₁ input ports on its input side, with each input port being a 1×n₂ wavelength demultiplexer, and N₂=r₂n₂ output ports on its output side, with each output port being an n₁×1 wavelength multiplexer, m_A r₁×r₂ AWGs in the central stage, where r₁=r₂=r, n₁=N₁/r, n₂=N₂/r, and m_A=n₁n₂;

In the AWG network S_A, the ith input port is labeled by D_A1(α_A,a_A), where α_A=└i/n₁┘ and α_A=[i]_n1, and the jth output port is labeled by M_A1(β_A,b_A), where β_A└j/n₂┘ and b_A=[j]_n2, and each of the AWGs in the central stage is labeled by G_A1(α_A,b_A); the α_Ath input port of G_A1(α_A,b_A) is connected with the b_Ath output port of D_A1(α_A,a_A), the β_A th output port of G_A1(α_A,b_A) is connected with the α_Ath input port of M_A1(β_A,b_A), and G_A1(α_A,b_A) is associated with a wavelength subset Λ_[αλ+bλ]n1={λ_{[αλ+bλ]n1r+k}|k=0, 1, . . . , r−1};

Step 5: substituting the ith N₂×1 wavelength multiplexer L_i and the 1×n₂ wavelength demultiplexer D_A1(α_A,a_A) with n₂ r×1 wavelength multiplexers, each of which is labeled by D_A2(α_A,a_A,b_A);

substituting the jth 1×N₁ wavelength demultiplexer R_j and the n₁×1 wavelength multiplexer M_A1(β_A,b_A) with n r×1 wavelength demultiplexers, each of the r×1 wavelength demultiplexers being labeled by M_A2(β_A,b_A,a_A);

associating each r₁×r₂ AWG, labeled by D_A2(α_A,b_A), with a wavelength subset (λ_k|k=0, 1, . . . , r−1);

where the output port of D_A2(α_A,a_A,b_A) is connected with the α_Ath input port of G_A2(a_A,b_A), and the input port of M_A2(β_A,b_A,a_A) is connected with the β_Ath output port of G_A2(α_A,b_A).

Alternatively, the method further comprises:

Step 4: in the case of N₁=KN₂, substituting the N₁×N₂ AWG with a two-stage network S_B(K,N₂,K,1,N₂), where each input port of the AWG network S_B is a link, there are K N₂×N₂ AWGs in the central stage, and each output port is a K×1 wavelength multiplexer; the ith input port is labeled by D_B1(α_B,a_B), where α_B=└i/N₂┘ and a_B=[i]_N2, the jth output port is labeled by M_B1(γ_B), where γ_B=j, and each AWG in the central stage is labeled by G_B1(α_B), and further, the AWG in the central stage is associated with a wavelength subset Λ_αB={λ_αBN2+k|k=0, 1, . . . , N₂−1}; the a_Bth input port of G_B1(α_B) is thus D_B1(α_B,a_B), the γ_Bth output port of G_B1(α_B) is connected with the α_Bth input port of M_B1(γ_B);

Step 5: denoting each of the N₁×1 wavelength multiplexers as D_B2(α_B,a_B);

Substituting the jth 1×N₁ demultiplexer R_j and the K×1 multiplexer M_B1(γ_B) with K N₂×1 wavelength multiplexers, each of which is labeled by M_B2(γ_B,α_B);

associating each N₂×N₂ AWG, labeled by G_B2(α_B), with a wavelength subset {λ_k|k=0, 1, . . . , N₂−1};

where D_B2(α_B,a_B) is connected with the a_Bth input port of G_B2(α_B), and M_B2(γ_B,α_B) is connected with the γ_Bth output port of G_B2(α_B).

The method further comprises:

Step 6: in the case of the dimension of the AWG in the central stage still being large, returning to Step 4, and substituting the AWG in the central stage with a module constituted of a three-stage network of small AWGs.

The present invention further provides an application of AWGs in an interconnection network. Compared with the previous works, the present invention has the following advantages:

(1) By constructing the N₁×N₂ interconnection network with the r×r AWGs, the number of the interconnection links is reduced r times.

(2) Wavelengths are reused in the AWG-based interconnection network, i.e., the r×r AWGs in the network reuses the same wavelength subset {λ₀, . . . , λ_r−1}, which improves the scalability of the AWG-based interconnection system.

(3) If r is still very large, the method in Step 4 can be employed to decompose the r×r AWG into an r×r three-stage AWG network.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 shows an N₁×N₂ interconnection network ₁ of the present invention;

FIG. 2 shows an AWG-based N₁×N₂ interconnection network ₂ of the present invention;

FIG. 3 shows an N₁×N₂ interconnection network _A1 comprising a three-stage AWG network S_A of the present invention;

FIG. 4 shows a KN₂×N₂ interconnection network _B1 comprising a two-stage AWG network S_B of the present invention;

FIG. 5 shows an N₁×N₂ AWG-based modular interconnection network _A2 of the present invention;

FIG. 6 shows a KN₂×N₂ AWG-based modular interconnection network _B2 of the present invention;

FIG. 7 shows a 15×10 interconnection network ₁ of the present invention;

FIG. 8 shows a 15×10 AWG-based interconnection network ₂ of the present invention;

FIG. 9 shows a 15×10 interconnection network _A1 comprising an AWG network S_A of the present invention;

FIG. 10 shows a 15×10 AWG-based modular interconnection network _A2 of the present invention;

FIG. 11 shows a 12×6 interconnection network ₁ of the present invention;

FIG. 12 shows a 12×6 AWG-based interconnection network ₂ of the present invention;

FIG. 13 shows a 12×6 AWG-based interconnection network _B1 comprising a two-stage AWG network S_B of the present invention;

FIG. 14 shows a 12×6 AWG-based modular interconnection network _B2 of the present invention.

DETAILED DESCRIPTION OF THE INVENTION AND PREFERRED EMBODIMENTS

In combination with drawings and embodiments hereunder provided, the present invention will be further expounded. The embodiments are intended as illustrating the present invention rather than limiting its scope. Various equivalent modifications by a person skilled in the art shall fall within the scope of the claims.

An N₁×N₂ (N₁≧N₂, and N₁ and N₂ having a maximum divisor r) interconnection network ₁ contains N₁ left nodes, with each node comprising N₂ ports, and each port having an optical transceiver; N₂ right nodes, with each node comprising N₁ ports, and each port having an optical transceiver. All the optical transceivers are associated with the same wavelength λ₀. For each of the paired left node and right node, there is one and only one link between them, and thus, there are altogether N₁N₂ links in the interconnection network, as shown in FIG. 1, where the small block on each port of the node represents an optical transceiver.

An AWG-based modular interconnection network for reducing the cabling complexity of a data center mainly via the wavelength division multiplexing property of AWGs, comprises the following parameters:

N₁×N₂ AWG: having N₁ input ports, N₂ output ports, and being associated with a wavelength set Λ={λ₀, λ₁, . . . , λ_N1−1}, as shown in FIG. 2;

N₂×N₂ AWG: having N₂ input ports, N₂ output ports, and being associated with a wavelength set Λ={λ₀, λ₁, . . . , λ_N2−1};

N₁×1 wavelength multiplexer/demultiplexer: N₁×1 AWG;

N₂×1 wavelength multiplexer/demultiplexer: N₂×1 AWG;

n₁×1 wavelength multiplexer/demultiplexer: n₁×1 AWG;

n₂×1 wavelength multiplexer/demultiplexer: n₂×1 AWG;

N₁×N₂ AWG network S_A(n₁, r₁, n₁, n₂, r₂): a three-stage AWG network, which comprises N₁=r₁n₁ input ports on the input side, each input port being a 1×n₂ demultiplexer, N₂=r₂n₂ output ports on the output side, each output port being an n₁×1 multiplexer, and m_A r₁×r₂ AWGs on the central stage, where r₁=r₂=r, n₁=N₁/r, n₂=N₂/r, and m_A=n₁n₂, as shown in FIG. 3;

KN₂×N₂ AWG network S_B(K, N₂, K, 1, N₂): an AWG network, with K N₂×N₂ AWGs in the first stage, and N₂ K×1 multiplexers in the second stage, as shown in FIG. 4;

N₁×N₂ interconnection network ₂:N₁ left nodes and N₁ N₂×1 multiplexers on the left side, N₂ right nodes and N₂ 1×N₁ demultiplexers on the right side, and an N₁×N₂ AWG on the central stage;

N₁×N₂ interconnection network _A1: consists of N₁ left nodes and N₁ N₂×1 multiplexers, N₂ right nodes and N₂ 1×N₁ demultiplexers, and an N₁×N₂ AWG network S_A on the central stage, as shown in FIG. 3;

N₁×N₂ interconnection network _B1: consists of N₁ left nodes and N₁ N₂×1 multiplexers; N₂ right nodes and N₂ 1×N₁ demultiplexers, and an N₁×N₂ AWG network S_B on the central stage, as shown in FIG. 4;

N₁×N₂ interconnection network _A2: consists of N₁ left nodes and N₁n₂r×1 multiplexers, N₂ right nodes and N₂n₁1×r demultiplexers, and n₁n₂ identical r×r AWGs on the central stage, as shown in FIG. 5;

N₁×N₂ interconnection network _B2: consists of N₁ left nodes and N₁ N₂×1 multiplexers, N₂ right nodes and KN₂ 1×N₁ demultiplexers, and K identical N₂×N₂ AWGs on the central stage, as shown in FIG. 6.

A method of construction of an AWG-based modular interconnection network comprises the following steps:

(1) Remove all the N₁N₂ cables in the N₁×N₂ interconnection network ₁ as shown in FIG. 1, then provide N₁ N₂×1 wavelength multiplexers, labeled by L₀, L₁, . . . , L_N1−1 for N₁ left nodes, the ith N₂×1 wavelength multiplexer having its jth input port connected with the jth port of the ith left node, and the jth port of the ith left node is associated with wavelength λ_[i+j]N1, where i=0, 1, . . . , N₁−1, and j=0, 1, . . . , N₂−1; provide N₂ 1×N₁ wavelength demultiplexers labeled by R₀, R₁, . . . , R_N2-1, for N₂ right nodes, the jth 1×N₁ wavelength demultiplexer having its ith output port connected separately with the ith port of the jth right node, and the ith port of the jth right node is associated with wavelength λ_[i+r]N1. Use N₁×N₂ AWG to connect N₁ N₂×1 wavelength multiplexers on the left with N₂1×N₁ wavelength demultiplexers on the right. This step transforms the ₁ into an ₂, with all the optical fibers being replaced by an AWG, as shown in FIG. 2, where the ith left node connects with the jth right node through wavelength λ_[i+j]N1.

As the fabrication of AWGs with very large N₁ and N₂ is impractical, it is necessary to decompose the AWG into a network of small AWGs: substitute the N₁×N₂ AWG with an AWG network S_A(n₁,r₁,m_A,n₂,r₂), as shown in FIG. 3. In S_A, each input port is a 1×n₂ demultiplexer and the ith input port is labeled by D_A1(α_A,a_A), where α_A=└i/n₁┘ and a_A=[i]_n1, and each output port is a n₁×1 multiplexer and the jth output port is labeled by M_A1(β_A,b_A), where β_A=└j/n₂┘, and b_A=[j]_n2, and an AWG in the central stage is labeled by G_A1(a_A,b_A); the α_Ath port of G_A1(a_A,b_A) is connected with the bath port of D_A1(α_A,a_A), the β_Ath port of G_A1(a_A,b_A) is connected with the a_Ath port of M_A1(β_A,b_A), and G_A1(a_A,b_A) is associated with a wavelength subset Λ_[aA+bA]n1={λ_[aA+bA]n1r+k|k=0, 1, . . . , r−1}. This step transforms ₂ to _A1, in which the ith left node connects with the jth right node via wavelength λ_{[aA+bA]n1r+[αA+βA]r} along the path as follows:

• Left Node i→output b_A of D_A1(α_A,a_A)→input α_A of G_A1(a_A,b_A)→output β_A of G_A1(a_A,b_A)→input a_A of M_A1(β_A,b_A)→Right Node j.

As shown in FIG. 4, when N₁=KN₂, the three-stage network S_A degenerates into a two-stage network S_B (K,N₂,K,1,N₂), where each input port degenerates into a 1×1 demultiplexer, i.e., a fiber link, the dimension of the AWGs in the central stage is N₂×N₂, and each output port becomes a K×1 multiplexer. The ith input port is labeled by D_B1(α_B,a_B), where α_B=└i/N₂┘, and a_B=[i]_N2, the jth output port is labeled by M_B1(γ_B), where γ_B=j, and an AWG in the central stage is labeled by G_B1(α_B), and further, an AWG in the central stage is associated with a wavelength subset Λ_αB={λ_αBN2+k|k=0, 1, . . . , N₂−1}; the a_Bth input port of G_B1(α_B) is thus D_B1(α_B,a_B), and the γ_Bth output port of G_B1(α_H) is connected with the α_Bth input port of M_B1(γ_B). This step transforms ₂ into _B1, where the ith left node connects with the jth right node via wavelength λ_{αBN2+[αB+γB]N2} along the path as follows:

• • Left Node i→D_B1(α_B,a_B)→input a_B of G_B1(α_B)→output γ_B of G_B1(α_B)→input α_B of M_B1(γ_B)→Right Node j.

Wavelengths are precious resources in an optical communication window, and the number of wavelengths required by the system should not increase with the dimension of the interconnection network, and therefore, wavelength reuse must be taken into account. As shown in FIG. 5, substitute the ith N₂×1 multiplexer L_i on the left side of _A1 and the 1×n₂ demultiplexer D_B1(α_B,a_B) of S_A with n₂ r×1 multiplexers, each of which is labeled by D_A2(α_A,a_A,b_A), and substitute the jth 1×N₁ demultiplexer R_j on the right side of _A1 and the n₁×1 multiplexer M_A1(β_A,b_A) of S_A with n₁ 1×r demultiplexers, each of which is labeled by M_A1(β_A,b_A,a_A). As the optical transceivers in the left and right nodes can perform wavelength isolation, each r×r AWG can reuse the same wavelength subset {λ_k|k=0, 1, . . . , r−1}. The r×r AWG in the central stage is labeled by G_A2(a_A,b_A). The output port of D_A2(α_A,a_A,b_A) is connected with the α_Ath input port of G_A2(a_A,b_A), and the input port of M_A2(β_A,b_A,a_A) is connected with the β_Ath output port of G_A2(a_A,b_A). Accordingly _A1 is transformed into _A2, where the ith left node connects with the jth right node via wavelength λ_[λαA+βA]r, along the path as follows:

• • Left Node i→D_A2(α_A,a_A,b_A)→input α_A of G_A2(α_A,b_A)→output β_A of G_A2(a_A,b_A)→M_A1(β_A,b_A,a_A)→Right Node j.

As shown in FIG. 6, if the outcome of step (2) is _B1, substitute the ith 1×N₁ demultiplexer R_j on the right side of _B1 and the jth K×1 multiplexer M_B1(γ_B) on the output side of S_B with K N₂×1 demultiplexers, each of which is labeled by M_B2(γ_B,α_B). Each N₂×N₂ AWG in the central stage is labeled by G_B2(α_B) and can be associated with the same wavelength subset {λ_k|k=0, 1, . . . , N₂−1}. The input port is labeled by D_B2(α_B,a_B). D_B2(α_B,a_B) is connected with the a_Bth input port of G_B2(α_B), and M_B2(γ_B,α_B) is connected with the γ_Bth output port of G_B2(α_B). As a result, _B1 is transformed into _B2, in which the ith left node connects with the jth right node via wavelength λ_[αB+γB]N2 along the path as follows:

• Left Node i→D_B2(α_B,a_B)→input a_B of G_B2(α_B)→output γ_B of G_B2(α_B)→M_B2(γ_B,α_B)→Right Node j.

(2) If the dimension of the AWG in the central stage is still very large, the method of step (2) can be employed to substitute each AWG in the central stage with a module encapsulated with a three-stage AWG network.

Example 1

A method for constructing an AWG-based modular interconnection network is shown. As for a 15×10 interconnection network ₁, as shown in FIG. 7, where the greatest common divisor of 15 and 10 is 5, the construction method of an AWG-based modular interconnection network comprises the following steps:

1. Substitute the 150 cables on FIG. 7 with a 15×10 AWG, which is associated with Λ={λ₀, λ₁, . . . , λ₁₄). ₁ is thus transformed into ₂. Compared with ₁, ₂ does not need optical fiber links, as shown in FIG. 8;

2. Decompose the 15×10 AWG in ₂ into an AWG network S_A(3,5,6,2,5), which consists of fifteen (15) 1×2 demultiplexers, six (6) 5×5 AWGs, and ten (10) 3×1 multiplexers. Accordingly, the wavelength set Λ=(λ₀, λ₁, . . . , λ₁₄} is divided to 3 subsets Λ₀={λ₀, . . . , λ₄}, Λ₁={λ₅, . . . , λ₉}, and Λ₂={λ₁₀, . . . , λ₁₄}, which are respectively associated with the six (6) 5×5 AWGs in the central stage. As shown in FIG. 9, this step transforms ₂ into _A1. Compared with the 150 optical fiber links of ₁, _A1 needs only 60 optical fiber links. The reduction ratio of the number of required optical fiber links in this example is 2.5;

3. Substitute a 1×2 demultiplexer of S_A and a 10×1 multiplexer on the left side of ₂ with two (2) 5×1 AWGs, and substitute a 3×1 multiplexer of S_A and the 1×15 demultiplexer on the right side of ₂ with three (3) 1×5 AWG Wavelength dependence among the 6 5×5 AWGs is eliminated due to the transceivers equipped on both sides of ₂. Thus, these AWGs can be associated with the same wavelength subset Λ₀, which indicates that the wavelength reuse property is achieved. As shown in FIG. 10, an AWG-based modular interconnection network ₂ is obtained. The number of required optical fibers in _A2 is 60. After the construction of an AWG-based modular interconnection network, the number of optical fiber links is reduced from 150 to 60, and the reduction of the number of optical fiber links remarkably cuts down the network complexity and thus simplifies system maintenance. In general, the number of optical fiber links is reduced r/2 times. In practical applications, N₁ and N₂ might be very large, which results in that the dimension of the AWGs in the central stage is still large. In this case, the method in step (2) can be employed to substitute each AWG in the central stage with a module that consists of a three-stage network of small AWGs.

Example 2

A method for constructing an AWG-based modular interconnection network, as for a 12×6 interconnection network ₁ as shown in FIG. 11, where 12 is a multiple of 6, the construction comprises the following steps:

1. Substitute the 72 linking cables in the middle of FIG. 11 with a 12×6 AWG, which is associated with Λ={λ₀, λ₁, . . . , λ₁₁}. This operation transforms ₁ to ₂. Compared with ₁, ₂ does not need optical fiber links, as shown in FIG. 12.

2. Decompose the 12×6 AWG in N₂ into an AWG network S_B(2,6,2,1,6) that consists of six (6) 1×2 multiplexers, two (2) 6×6 AWGs. Accordingly, Λ={λ₀, λ₁, . . . , λ₁₄} is divided into 2 subsets Λ₀={λ₀, . . . , λ₅} and Λ₁={λ₆, . . . , λ₁₁}, which are respectively associated with the two (2) 6×6 AWGs in the central stage. As shown in FIG. 13, this step transforms into ₂ Compared with the optical fiber links of ₁, _B1 needs only 12 optical fiber links. The reduction ratio of the number of optical fiber links in this example is 6.

3. Substitute a 2×1 multiplexer of S_B and a 1×12 demultiplexer on the right side of ₂ with two (2) 1×6 AWGs. Wavelength dependence between the two (2) 6×6 AWGs is eliminated due to the transceivers provided on both sides of ₂. Thus, these AWGs can be associated with the same wavelength subset Λ₀, which indicates that the wavelength reuse property is achieved. As shown in FIG. 14, an AWG-based modular interconnection network _B2 is obtained. The number of required optical fibers in _B2 is 12. In practical applications, N₁ and N₂ might become very large, resulting in large AWGs in the central stage. In such a case, the method in step (2) can be employed to substitute the AWG in the central stage with a module that consists of a three-stage network of small AWGs. This is the idea of modulization of the present invention. After the construction of an AWG-based modular interconnection network, the number of optical fiber links is reduced from 72 to 12, a 6 time reduction. General speaking, when N₁ is a multiple of N₂, the number of optical fiber links can be reduced from N₁N₂ to N₁, by employing the method. The reduction of the number of optical fiber links cuts down the network complexity and thus simplifies network maintenance.

Claims

1. A method for constructing an interconnection network, comprising: providing N1 N2×1 wavelength multiplexers, labeled by L0, L1, . . . , LN_−1, for N1 left nodes, an ith N2×1 wavelength multiplexer having its jth input port connected with a jth port of an ith left node, and a jth port of the ith left node is associated with a wavelength

2. An AWG-based modular interconnection network according to claim 1, comprising: left nodes, a number of the left nodes being N1, and each of the left nodes having N2 ports,right nodes, a number of the right nodes being N2, and each of the right nodes having N1 ports,an optical transceiver associated with a fixed wavelength on each port of the left and right nodes,r×1 wavelength multiplexers, a number of the r×1 wavelength multiplexers being N1n2, each of the n2 r×1 wavelength multiplexers having N2 input ports being connected with the N2 ports of one of the left nodes, and n2=N2/r,1×r wavelength demultiplexers, a number of the 1×r wavelength demultiplexers being N2n1, each of the n1 1×r wavelength demultiplexers having N1 output ports being connected with the N1 ports of one of the right nodes, and n1=N1/r, andr×r AWGs connecting the r×1 wavelength multiplexers and the 1×r wavelength demultiplexer, a number of the r×r AWGs being n1n2, each of the r×r AWGs being associated with a wavelength subset {λk|k=0, 1, . . . , r−1},where N1≧N2, N1 and N2 are integers that have a greatest common divisor r, and r>1.

3. The method as claimed in claim 1, further comprising: replacing the AWG in the central stage with a module constituted of a network AWG of three stages.

4. A method for constructing an interconnection network, comprising: providing N1 N2×1 wavelength multiplexers, labeled by L0, L1, . . . , LN1−1, for N1 left nodes, an ith N2×1 wavelength multiplexer having its ith input port connected with a jth port of an ith left node, and a jth port of the ith left node is associated with a wavelength

5. An AWG-based modular interconnection network according to claim 4, comprising: left nodes, a number of the left nodes being N1, and each of the left nodes having N2 ports,right nodes, a number of the right nodes being N2, and each of the right nodes having N1 ports,an optical transceiver associated with a fixed wavelength on each port of the left or right node,N2×1 wavelength multiplexers, a number of the N2×1 wavelength multiplexers being N1, and each of the N2×1 wavelength multiplexers having N2 input ports being connected with the N2 ports of one of the left nodes;1×N2 wavelength demultiplexers, a number of the 1×Nz wavelength demultiplexers being N1=KN2, and each of the K 1×N2 wavelength demultiplexers having N1 output ports being connected with the N1 ports of one of the right nodes; andN2×N2 AWGs connecting the N2×1 wavelength multiplexers and the 1×N2 wavelength demultiplexers, a number of the N2×N2 AWGs being K, and each of the N2×N2 AWGs being associated with a wavelength subset {λi|i=0, 1, . . . , N2−1},wherein N1=KN2.

6. The method as claimed in claim 4, further comprising: replacing the AWG in the central stage with a module constituted of a three-stage network of AWGs.