The present invention relates to a communication system and a communication method, more particularly to a communication system with a multi-grade fabric configuration and a communication method for a multi-grade service.
In known technology, function separation of network devices and packet-transport integration have been promoted to combine multi-vendor apparatuses and configure a network. For example, Non Patent Literature (NPL) 1 below describes a slice network proposed as a method for building a flexible virtual network. In NPL 2 below, segment routing is defined based on tag information.
For recent communication systems, virtual networks (vNW) need to be built according to various requirements. For the physical resources that constitute the communication system, devices having uniform performance are generally used. When the physical resources are configured of devices having uniform performance, there is a problem in that physical resources are deployed according to the virtual network having the highest performance requirements. As a result, physical resources may be utilized inefficiently.
The communication system 30 includes a plurality of Spine switches (designated as “Spin SW” in the drawing) 302, a plurality of Leaf switches (designated as “Leaf SW” in the drawing) 304, and a server 306 in which a virtual machine (VM) is activated to execute an application.
The Spine switches 302 and the Leaf switches 304 illustrated in
In light of the foregoing, an object of the present invention is to efficiently utilize physical resources in a communication system that builds virtual networks having various requirements.
To achieve the object described above, a first aspect of the present invention provides a communication system includes: physical resources including a Spine switch group consisting of a plurality of Spine switches, a Leaf switch group consisting of a plurality of Leaf switches, and a plurality of servers connected to any one of the Leaf switches, and a controller configured to build a virtual network on the physical resources, in which at least one of the plurality of Spine switch group and the plurality of Leaf switch group is constituted by a mix of switch devices having different performance, and the controller selects a physical resource of the physical resources used to build the virtual network based on desired performance of the virtual network.
A fourth aspect of the present invention provides a communication method for building a virtual network on physical resources including a Spine switch group consisting of a plurality of Spine switches, a Leaf switch group consisting of a plurality of Leaf switches, and a plurality of servers connected to any one of the plurality of Leaf switches, the communication method including: selecting a physical resource of the physical resources to be used for building the virtual network based on desired performance of the virtual network, in which at least one of the Spine switch group and the Leaf switch group is constituted by a mix of switch devices having different performance.
With this configuration, physical resources to be used can be selected as appropriate according to the desired performance of the virtual network, and systems can be constructed more efficiently (at low cost) than when uniformly disposing all physical resources based on the virtual network with the highest performance requirements.
In addition, in the communication system according to a second aspect of the present invention the controller obtains performance information of each of the plurality of Spine switches and each of the plurality of Leaf switches, sequentially monitors an operational status of each of the plurality of Spine switches and each of the plurality of Leaf switches, and selects a Spine switch of the plurality of Spine switches, a Leaf switch of the plurality of Leaf switches, and a server of the plurality of servers to be used to build the virtual network.
The communication method according to fifth aspect of the present invention further includes obtaining performance information of each of the plurality of Spine switches and each of the plurality of Leaf switches, and monitoring an operational status of each of the plurality of Spine switches and each of the plurality of Leaf switches, in which, in the resource selection step, a Spine switch of the plurality of Spine switches, a Leaf switch of the plurality of Leaf switches, and a server of the plurality of servers to be used for building the virtual network are selected based on the information obtained in the performance information obtainment step.
With this configuration, appropriate physical resources can be selected according to the desired performance of the virtual network.
Furthermore, in the communication system according to a third aspect of the present invention, the Spine switch group is constituted by optical path switches, and the Leaf switch group is constituted by a mix of switch devices having different performance.
In the communication method according to the sixth aspect of the present invention, the Spine switch group is constituted by optical path switches, and the Leaf switch group is constituted by a mix of switch devices having different performance.
With this configuration, Leaf-to-Leaf traffic can be isolated and bandwidth can be secured to improve communication quality.
According to the present invention, physical resources can be efficiently utilized in a communication system that builds virtual networks having various requirements.
Hereinafter, preferred embodiments of a communication system and a communication method according to the present invention will be described in detail with reference to the accompanying drawings.
A communication system 10 according to the first embodiment includes a physical resource including a Spine switch group 12 consisting of a plurality of Spine switches (designated as “Spin SW” in the
The plurality of Spine switches 102 and the plurality of Leaf switches 104 are connected in full mesh topology.
To facilitate visibility in the drawings, connection lines are partially omitted. However, the controller 110 is connected to all switches (Spine switches 102 and Leaf switches 104) in the communication system 10.
Various types of virtual networks (see
Here, at least one of the Spine switch group 12 and the Leaf switch group 14 is constituted of a mix of switch devices having different performance. In other words, the communication system 10 has a multi-grade fabric configuration.
In the first embodiment, both the Spine switch group 12 and the Leaf switch group 14 are constituted of a mix of switch devices having different performance.
In the example of
In addition, the Leaf switch 104A is a virtual switch (vSW), the Leaf switch 104B and the Leaf switch 104D are white box switches (WBSW), and the Leaf switches 104C and 104E are black box switches (BBSW).
A network configured by devices having different performance as described above may form a path between switches having varying performance depending on the performance of the switches passing through the path.
Thus, the controller 110 selects physical resources to be used for building a virtual network based on the desired performance of the virtual network.
More specifically, the controller 110 obtains performance information of each Spine switch 102 and each Leaf switch 104. In addition, the controller 110 sequentially monitors the operational statuses of the Spine switches 102 and the Leaf switches 104 (including the link performance between the Spine switches 102 and the Leaf switches 104) and selects the Spine switch 102, the Leaf switch 104, and the server 106 to be used for building the virtual network.
Note that all of the locations corresponding to the numerical values in
The performance information of each switch device is static information determined based on the intrinsic performance of each device. The operational status of each switch device is variable information obtained by monitoring the performance of each switch device. Examples of the performance information of each switch device include data rate, data rate type, and reliability (availability).
Examples of the operational status of each switch device include packet loss rate, wait time, jitter, and survival time.
In addition, the controller 110 monitors the performance of the link connecting each Spine switch 102 and each Leaf switch 104.
Note that the items of performance information and operational status illustrated in
The controller 110 selects physical resources that match the request for the virtual network, and determines a communication path and the server 106 to execute the application. That is, the controller 110 selects an input Leaf switch port, a relay Spine switch, an output Leaf switch port, and an application execution server depending on the desired performance of the virtual network.
In
The Spine switch 102B, which is a white box switch (WBSW), is inexpensive and has high speed. The Spine switch 102D, which is an optical path switch (Opt Path SW), has low latency and high reliability. The Leaf switch 104A, which is a virtual switch (vSW), is inexpensive and has high speed. The Leaf switches 104C and 104E, which are black box switches (BBSW), have high speed and high reliability.
Note that the characteristics of these switches are the result of abstractly paraphrasing the performance information and operational status (numerical information) of each switch as illustrated in
Here, the user (user terminal) 120A has the need to utilize a virtual network with low latency and high reliability. The user (user terminal) 120B has the need to utilize a virtual network at low cost above all else. Virtual networks with types compatible with each of these needs are selected, and physical resources (communication paths) matching the desired performance of each virtual network (see
For example, for the user 120A, as indicated by the thick lines in
For the user 120B, as indicated by the dashed lines in
As illustrated in
For example, as indicated by the bold lines in
The controller 110 obtains the performance information (static information) of each switch device in the communication system 10 and puts the data into a database (DB) (Step S600: Performance information obtainment step).
Each switch device sequentially monitors the performance of a path connecting that switch device itself to another switch device (Step S610).
The controller 110 obtains path performance information from each switch device, monitors (obtains) the operational status of each switch device, and puts the data into a database with these pieces of dynamic information (Step S601: Performance information obtainment step).
The controller 110 is populated with the desired performance of each virtual network (vNW) and puts the data into a database of this information (Step S602).
The controller 110 compares the desired performance of each virtual network with the contents of the consequential databases of Steps S600 and S601, and then calculates candidates for the optimal path (the switch devices through which signals passes) and the server 106 in which the virtual machine (VM) is disposed (Step S603).
Then, in consideration of the current allocation state, the physical resource that is actually set as the path from the path candidates is selected, and a path is allocated (Step S604: Resource selection step).
The controller 110 updates the path information based on the result of allocation in Step S604 (Step S605). Then, the server 106 that activates the application function necessary for building the virtual network is determined (Step S606: Resource selection step).
Each switch device also embeds a tag of the path information determined in Step S604 in a packet (Step S611) and performs routing based on the tag information (Step S612).
As described above, with the communication system 10 according to the first embodiment, the physical resources to be used can be selected as appropriate based on the desired performance of the virtual network, and a system can be constructed more efficiently (at low cost) than when uniformly disposing all of the physical resources according to the virtual network with the highest performance requirements.
In the first embodiment, a case is described in which both the Spine switch group and the Leaf switch group are constituted of a mix of switch devices having different performance. The second embodiment deals with a case where the Spine switch group is constituted by only optical path switches and the Leaf switch group is constituted by a mix of switch devices having different performance.
In a communication system 20 according to the second embodiment illustrated in
In the communication system 20, the Spine switches 202 (202A and 202B, designated as “Opt Path SW” in the drawing) are both optical path switches (Opt Path SW). With the optical path switch, wavelength separation is performed on wavelength-multiplexed input signals output from each Spine switch 202 and wavelength routing is performed in accordance with a previously constructed path to multiplex the input signals and transmit data to the output side of each Spine switch. A switch other than an optical path switch is a layer 2 (L2) or a layer 3 (L3) switch, whereas an optical path switch is a layer 0 (L0) or a layer L1 (L1) switch, and is a switch underlying than other switches. The optical path switch switches the optical path itself, through which a layer 2 (L2) signal or a layer 3 (L3) signal transmitted from another other switch passes. An optical path switch cannot recognize itself from other switches.
The Leaf switches 204 (204A to 204D) are any type of switch device. The Leaf switches 204 are connected to wavelength multi/demultiplexer modules (Mux/Demux) 206 (206A to 206D). The wavelength multi/demultiplexer module 206 outputs a plurality of single wavelength signals having different wavelengths input from the Leaf switch 204 as one wavelength multiplexed (WDM) signal to the Spine switch 202 (Mux). The wavelength multi/demultiplexer module 206 then splits the one WDM signal input from the Spine switch 202 into a plurality of single wavelength signals and outputs these signals to the Leaf switch 204 (Demux).
For example, when signals having wavelengths of λ1, λ2, λ3 are input from the Leaf switch 204A to the wavelength multi/demultiplexer module 206A, the wavelength multi/demultiplexer module 206A outputs the signals having the wavelengths λ1, λ2, λ3 to the Spine switch 202A as one WDM signal. The Spine switch 202A uses wavelength routing to output the signal having the wavelength λ1 to a wavelength multi/demultiplexer module 206B (Leaf switch 204B), the signal having the wavelength λ2 to a wavelength multi/demultiplexer module 206C (Leaf switch 204C), and the signal having the wavelength λ3 to a wavelength multi/demultiplexer module 206D (Leaf switch 204D).
By using an optical path switch as the Spine switch 202, separating (isolating) the wavelength for each requirement when providing networks for diverse traffic with different requirements can generate independent bands between wavelengths. This not only provides economic paths that are statistically multiplexed as in known technology, but also generates completely independent high quality signals that ensure bands and not affect other traffic.
In other words, in a known Clos topology, Leafs are not directly connected to each other for horizontal traffic exchange between Leafs, and traffic flows through Spines in an upper hierarchy to other Leafs. This leads to problems such as poor efficiency, reduced performance, and band compression.
In contrast, with the configuration according to the present invention illustrated in
An example of an optical path switch described above is an Arrayed Waveguide Grating (AWG) router. The AWG router builds a full mesh network using cyclic AWG. More specifically, a full-mesh network can be built with star-shaped optical fibers, which can greatly reduce the number of fiber optics. In addition, since there is no need to mutually convert light and electricity, no power supply is necessary which makes the network highly reliable.
Further, bandwidth can be expanded without adding Spine switches 202 by using the transmission characteristics of a wavelength router such as the AWG router to insert multiple signals into wavelength paths on the same path. Specifically, the following methods are conceivable.
Method 1
Multiple signals are routed to the same path by passing multiple signals through the same transmission band.
For example, in
If bandwidth expansion is required in such a state (e.g., if an additional signal S is output from the Leaf switch 204), a signal having the wavelength λ4 included in the filter bandwidth WB2 (additional signal) is configured to be output from the output port 3. Similarly, a signal having the wavelength λ2 included in the filter bandwidth WB1 (indicated by the dotted line) is configured to be output from the output port 2. Further, the signal having the wavelength λ6 included in the filter bandwidth WB3 (indicated by the dotted line) is configured to be output from the output port 4. That is, the Method 1 is a method of performing bandwidth expansion by causing a plurality of signals to pass through one wavelength band filter.
Method 2
Individual signals are passed through a plurality of bands that perform the same routing to route multiple signals to the same path.
For example, in
If bandwidth expansion is required in such a state, for example, a filter (filter bandwidth=λ5) is set to the output port 3 such that a signal having the wavelength λ5 is output, causing the signal having a wavelength λ5 (additional signal) to be output in addition to a signal having the wavelength λ2. Similarly, a filter (filter bandwidth=λ4) is set to the output port 2 such that a signal having the wavelength λ4 is output, causing the signal having the wavelength λ4 (indicated by the dotted line) to be output in addition to a signal having the wavelength λ1. In addition, in the output port 4, a filter (filter bandwidth=λ6) is set to the output port 4 such that a signal having the wavelength λ6 is output, causing a signal having a wavelength λ6 (indicated by a dotted line) to be output in addition to a signal having the wavelength λ3. That is, Method 2 is a method of performing band expansion by setting a plurality of wavelength band filters for one output port.
Method 3
Method 1 and Method 2 can also be used in combination.
Further, redundancy of the Spine switch 202 can be achieved without adding a port on the Leaf switch 204 side, for example.
For example, as illustrated in
In terms of a cluster-to-cluster connection configuration, the Spine router does not have a switch function, which results in a configuration that connects Borders and Leafs (B-Leaf) to each other.
While configuring a Spine-to-Spine connection model is not possible, a B-Leaf-to-B-Leaf connection model is superior to a Spine-to-Spine connection model in terms of scalability, redundancy-cluster operation efficiency, and transition from single clusters to multiple clusters. Further, B-Leaf-to-B-Leaf connection is common in view of city trends.
The performance of the communication system according to the second embodiment (designated as the present invention) and the performance of known configurations are discussed in comparison.
The known configurations to be compared are the Clos network illustrated in
Traffic Quality Between Leafs
In a Clos network, bandwidth between Leafs is shared, but the network is strong against fluidity and has efficient accommodation through statistical multiplexing. A Fullmesh network has two types, that is, a type where Leafs are directly connected to each other and a type where statistical multiplexing is shared via the Spines. As a result, accommodation can be made efficient by understanding and designing cross traffic in network. In contrast, in the present invention, Leaf-to-Leaf traffic can be isolated and bandwidth can be secured such that this traffic is not affected by traffic on other paths and communication quality is further improved. Low latency transfer is also possible. That is, in terms of Leaf-to-Leaf traffic quality, the present invention is advantageous over known configurations of Clos networks and Fullmesh networks.
Failure Resistance
A Clos network is capable of switching only fault sites, and only traffic accommodated by the fault sites is affected. The same applies to a Fullmesh network, where only fault sites can be switched and only traffic accommodated by the fault sites is affected. In contrast, while in the present invention only fault sites can be switched and only traffic accommodated by the fault sites is affected similarly to the above network types, in terms of failure resistance, the present invention achieves higher Spine reliability and higher reliability of clusters than Clos networks and Fullmesh networks, and is therefore advantageous over known configurations.
Operability
In a Clos network, it is difficult to secure routes when balancing load. In a Fullmesh network, it is difficult to secure routes when balancing load, but direct connection between Leafs can be identified. In contrast, in the present invention, Leafs appear to be directly connected to each other in a full mesh topology in layers L2 and above, and this configuration is easy to control because P2P connection is established. In other words, the present invention is advantageous over known configurations of Clos networks and Fullmesh networks.
Considering an effective use case of the present invention, the present invention is effective in cases where low latency transfer is required, or when Leaf-to-Leaf traffic is often fixed. For example, the present invention is considered effective in a small data center (DC) such as a data center interconnect (DCI) or in a metropolitan area such as a capital city because scalability in terms of number of ports is low and costs are low.
Number | Date | Country | Kind |
---|---|---|---|
2018-152241 | Aug 2018 | JP | national |
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/JP2019/031074 | 8/7/2019 | WO | 00 |