CN109302350B

CN109302350B - Scheduling method based on photoelectric multicast hybrid network frame structure

Info

Publication number: CN109302350B
Application number: CN201710609953.1A
Authority: CN
Inventors: 赵宝康; 董德尊; 刘波; 保金帧; 赵静月; 王宝生; 陶静; 时向泉; 毛席龙; 虞万荣; 原玉磊; 韩彪
Original assignee: National University of Defense Technology
Current assignee: National University of Defense Technology
Priority date: 2017-07-25
Filing date: 2017-07-25
Publication date: 2021-05-11
Anticipated expiration: 2037-07-25
Also published as: CN109302350A

Abstract

The invention relates to the field of multicast communication, in particular to a hybrid network frame structure based on photoelectric multicast, which comprises: EPS, OCS, SDN controller, TOR switch, server, the TOR switch with EPS, OCS and server are connected, the SDN controller with the switch, the EPS and the OCS are connected. The framework avoids the influence of optical link configuration on the multicast stream and reduces the optical link configuration overhead by seamlessly scheduling the multicast stream between the static electric network and the dynamic optical network; the framework integrates the characteristics of low power consumption of optical multicast and direct connection of the TOR, and also has the advantage of low switching delay of electric multicast.

Description

Scheduling method based on photoelectric multicast hybrid network frame structure

Technical Field

The invention relates to the field of communication, in particular to a hybrid network frame structure based on photoelectric multicast and a scheduling method thereof.

Background

The current multicast solutions for high performance data center applications mainly include application layer overlay networks and network layer multicast.

Application layer overlay network solutions speed up high-volume data transmission by building a data distribution network on the overlay network, where the receiving node also participates in the sending of data, sending the newly received data to the next node along the distribution network. The method avoids the problem of high load of a single sending node, and accelerates the transmission of multicast data; however, the method introduces redundant traffic, and the unbalanced arrival between nodes also affects the performance of data distribution, and cannot achieve the optimal multicast optimization effect.

Network layer multicast enables a source node to send only one copy of data, which is sent along the multicast tree to all recipient nodes. The message is copied and distributed in multiple copies when passing through the OCS or the EPS. The traditional Ethernet multicast protocol builds a multicast tree through distributed negotiation, and the building of the multicast tree and the management of multicast group members are complex. Avalanche manages multicast groups through SDN controllers and constructs multicast trees in the core electrical packet switch network. The EPS network usually adopts a tree network topology, in the core EPS, the unicast stream and the multicast stream coexist, and the multicast stream competes for the link bandwidth with the unicast stream passing through the core EPS, thereby reducing the throughput of the multicast stream and the unicast stream and increasing the transmission time of the streams. In order to eliminate the contention of the core EPS, someone uses the OCS to copy the optical links, and one to many optical links are established between the source and destination counter top switches to migrate the multicast traffic to the optical network. However, the optical link needs to be allocated to the multicast stream in advance, and the multicast stream is monopolized, and the multicast stream that is not allocated needs to wait for the available optical link resource, thereby reducing the performance of the multicast stream. The OCS configuration requires a certain delay, usually in milliseconds, and the current solution does not consider the influence of the configuration delay, thereby affecting the performance of the multicast stream.

Disclosure of Invention

The invention aims to provide a hybrid network frame structure based on photoelectric multicast and a scheduling method thereof, which solve the problem of multicast optimization effect.

The invention is realized in this way, a structure and its scheduling method based on mixed network frame of photoelectricity multicasting, the structure includes: EPS, OCS, SDN controller, TOR switch, server, the TOR switch with EPS, OCS and server are connected, the SDN controller with the switch, the EPS and the OCS are connected.

The further technical scheme of the invention is as follows: the SDN controller comprises: an application layer plane, a control plane, and a data plane.

The further technical scheme of the invention is as follows: the method comprises the following steps: a hybrid scheduling strategy is adopted, and the hybrid scheduling strategy comprises the following steps: a greedy link allocation method.

The further technical scheme of the invention is as follows: the hybrid scheduling policy is: when the application layer of the SDN controller senses that a new multicast flow arrives, an application program explicitly sends a request to the network architecture by calling a northbound interface; after receiving a group of multicast requests, firstly aggregating the traffic of the server level into the requirement of the TOR level, then executing a greedy optical link allocation algorithm, calculating the OCS configuration condition, and searching for an available optical switch port; because a period of time is required to be occupied during OCS configuration, a multicast tree is constructed for all multicast streams according to an EPS network, and then the multicast tree is issued to a routing table of the TOR switch through a southbound interface, so that the multicast streams can be transmitted through the EPS network, and the phenomenon that the transmission is suspended in the process to wait for available OCS resources or the OCS configuration is completed is avoided; and after the OCS configuration is completed, calculating a new multicast tree and updating a routing table, transferring the multicast stream which is distributed with the optical link to the OCS network, and completing seamless switching of the multicast stream in the optical network and the electric network.

The further technical scheme of the invention is as follows: the greedy link allocation method comprises the following steps: and sequencing the multicast streams from large message volume to small message volume, wherein the multicast streams with large message volume are preferentially met. For each multicast stream, an idle source TOR switch and a destination TOR switch that can be connected are found. If the idle exchanger can cover all members of the multicast group, the multicast flow can be distributed; to reduce power consumption, the algorithm employs OCS multicast as much as possible, i.e. selects a smaller number of EPS ports to cover the multicast group, and the algorithm is executed until all multicast streams can be allocated.

The invention has the beneficial effects that: in the process of optical link configuration, a plurality of optical/electrical multicast hybrid network architectures transmit all multicast streams on an EPS network, and when the optical link configuration is completed, the multicast streams are switched to an OCS network for transmission; the framework avoids the influence of optical link configuration on the multicast stream and reduces the optical link configuration overhead by seamlessly scheduling the multicast stream between the static electric network and the dynamic optical network; the framework integrates the characteristics of low power consumption of optical multicast and direct connection of the TOR, and also has the advantage of low switching delay of electric multicast.

Drawings

Fig. 1 is a schematic structural diagram of a hybrid network architecture based on photoelectric multicast according to the present invention;

fig. 2 is a schematic flow chart of a hybrid scheduling policy of a scheduling method based on a hybrid network architecture for optical-electrical multicast according to the present invention;

fig. 3 is a schematic flow diagram of a greedy link allocation method of a scheduling method based on a hybrid optical/electrical multicast network architecture provided in the present invention.

Reference numerals: 1. SDN controller 2, EPS 3, OCS 4, connecting line 5, TOR switch 6, server.

Detailed Description

The first embodiment is as follows: fig. 1 shows a hybrid network architecture based on opto-electrical multicast, which is composed of EPS2, OCS3, SDN controller 1, TOR switch 5, and server 6; the TOR switch 5 is connected with the EPS2 and the OCS3 through a connecting line 4 and is simultaneously connected with a plurality of servers 6; where OCS2 is equipped with a passive optical splitter, SDN controller 1 manages EPS2 and OCS 3. EPS2 and OCS3 support both unicast and multicast communications, EPS2 implements point-to-multipoint communications by replicating packets, OCS3 uses a passive optical splitter to split optical signals at the physical layer to establish a one-to-many connection to support optical layer multicast, and SDN controller 1 includes an application layer plane, a control plane and a data plane, the control plane maintaining full network attempts, including the state of electrical and optical switches. The controller receives the multicast demand matrix from the application plane through the northbound interface, then calculates the configuration and routing strategy of the optical switch, and configures the optical switch and the electric switch through the southbound interface after the calculation is completed. In order to eliminate the influence caused by the optical link switching delay, the control plane adopts a mixed scheduling strategy to seamlessly switch the multicast stream in the optical network and the electric network. Meanwhile, the control plane adopts a greedy link allocation algorithm for efficiently integrating optical multicast and electric multicast to allocate multicast resources and construct a multicast tree, so that multicast streams are effectively optimized.

Fig. 2 shows a hybrid scheduling strategy of a scheduling method based on a hybrid network architecture for optical-electrical multicast, where 1, a multicast stream demand matrix arrives; 2. the multicast demand matrix changes; 3. executing a greedy link allocation algorithm to allocate an OCS link for the multicast flow; 4. after calculating new link allocation, starting to configure the OCS, and in the process of configuring the OCS, transmitting the multicast stream on the EPS network by configuring the route of the TOR switch; 5. after OCS configuration is completed, the route of the TOR switch is configured, so that the multicast flow which is allocated with the optical link is transferred from the EPS network to the OCS network for transmission; 6. and completing the transmission of the multicast stream transmitted by the OCS.

Fig. 3 is a schematic flow diagram illustrating a greedy link allocation method of a scheduling method based on a hybrid multicast optical network architecture, where the greedy link allocation method includes: 1. configuring a given multicast flow demand matrix and the current OCS; 2. traversing all multicast streams from large to small according to the size of the data volume sent by the multicast streams; 3. for each multicast flow i, searching an idle port set on a corresponding source TOP switch and idle end sets on all destination TOR switches; 4. whether the intersection of all the destination TOR switch idle port sets belongs to the source TOR switch idle port set or not is judged, if yes, the step 5 is carried out, and if not, the step 3 is returned; 5. the minimum set of ports that can cover all destination TOR switches in multicast stream i is selected for allocation.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims

1. A scheduling method based on a photoelectric multicast hybrid network frame structure is characterized by comprising the following steps: a hybrid scheduling strategy is adopted, and the hybrid scheduling strategy comprises the following steps: a greedy link allocation method; the hybrid scheduling policy is: when an application layer of the SDN controller senses the arrival of a new multicast stream, an application program explicitly sends a request to the network architecture by calling a northbound interface; after receiving a group of multicast requests, aggregating the traffic of server level into the requirement of top of rack TOR level, then executing a greedy link allocation method, calculating the OCS configuration condition of the optical circuit switch, and searching for an available optical switch port; because the OCS needs to occupy a period of time during the configuration of the OCS, a multicast tree is constructed for all multicast streams according to the EPS network of the electric packet switch, and then the multicast tree is issued to the routing table of the top-of-rack TOR switch through a southbound interface, so that the multicast streams can be transmitted through the EPS network of the electric packet switch, and the phenomenon that the transmission is suspended in the process to wait for available OCS resources of the OCS or complete the configuration of the OCS is avoided; after the OCS configuration is completed, calculating a new multicast tree and updating a routing table, migrating the multicast stream which is distributed with the optical link to an OCS network, and completing seamless multicast stream switching in an optical network and an electric network; the greedy link allocation method comprises the following steps: the multicast streams are sequenced from large message volume to small message volume, and the multicast streams with large message volume are preferentially met; for each multicast stream, searching an idle source top TOR switch and an idle destination top TOR switch which can be connected; if the idle exchanger can cover all members of the multicast group, the multicast flow can be distributed; in order to reduce power consumption, the greedy link allocation method adopts OCS multicast, namely, selects fewer EPS ports of the electric packet switch to cover a multicast group, and is executed until all multicast streams can be allocated.