KR102028651B1 - Method for selecting shard leader for fast synchronizations in software defined network - Google Patents

Abstract

A method of selecting a shard leader is disclosed. The method of selecting a shard reader is performed by a leader selection module included in a software defined network system including a plurality of controllers, and the available synchronization bandwidth for all controllers, the average required synchronization bandwidth for all shards, And calculating a limit synchronization bandwidth for all controllers, if a controller having a negative value of the limit synchronization bandwidth is present, selecting a shard having a minimum average required synchronization bandwidth from among the shards in which the controller acts as a leader. Selecting a controller in which the limit synchronization bandwidth is not negative even if a new leader of the shard is selected; And transferring the leadership of the corresponding shard to the selected controller as a new leader and updating the leader mapping state.

Description

METHOD FOR SELECTING SHARD LEADER FOR FAST SYNCHRONIZATIONS IN SOFTWARE DEFINED NETWORK}

An embodiment according to the concept of the present invention relates to a method for selecting a leader for fast controller synchronization in a software defined network. More specifically, the present invention provides a method for selecting a leader for fast synchronization in a distributed software defined network environment including a plurality of controllers. The present invention relates to a shard leader selection method in consideration of network bandwidth and shard available network bandwidth requirements.

Software Defined Network (SDN) is proposed to fundamentally solve the high cost and complexity of network construction, management, and operation, and it is designed to replace the control part and packet (or flow) delivery part of existing network equipment. It has a structure to separate.

In a distributed SDN controller environment, multiple SDN controllers control a set of network devices (eg network switches) that are independent of each other. In addition, controllers share information (ie, databases) about networks that are managed independently from each other through distributed database technology. Specifically, the controller database is divided into a number of logical data partitions (hereinafter referred to as 'shards' or 'shards'), and replicas for each shard are stored in all controllers. Shard-specific replicas act as independent Raft groups. When a leader replica for any shard receives a data change request (e.g. create, update, delete), it sends it to other replicas (i.e. Follower Replicas) over the network for the data change according to the request. Perform the task of cloning. The shard leader confirms to the majority of follower replicas that the replication is complete and then reflects the data changes in the database. In addition, when another request for access (creation, update, read, delete) of the data arrives from the time the Leader Replica receives a change request for any data to the majority of Follower Replicas, the replication is completed. Requests are queued and processed sequentially when processing for previous requests (ie, replication and replication confirmation) is complete.

Meanwhile, when synchronization between SDN controllers is performed in a WAN environment as shown in FIG. 1, since available network bandwidth (ABW) is limited, the available network bandwidth from the controller where the leader is located to the controllers where the followers are located. Replication (ie, synchronization) bandwidth for that shard may be limited. In other words, if the available network bandwidth from the controller where the leader replica of any shard is located to the controller where the follower replicas are very low, the waiting time in the queue where the change requests for the data contained in the shard are stored may increase. This delays replication (synchronization), and as a result, access request throughput for data contained in the shard may be degraded.

Therefore, there is a need to arrange shard leaders in controllers in consideration of the available network bandwidth between the controllers so that the synchronization bandwidth of the shards is not limited by the network.

In addition, in the existing Raft Consensus Algorithm, the Leader is elected based on the Randomized Timeout without considering the available network bandwidth among the aforementioned controllers, which may result in deterioration of control plane performance due to synchronization delay.

Therefore, the present invention proposes a method of selecting a shard leader considering the available network bandwidth between controllers and the available network bandwidth requirement of the shard for fast synchronization.

US 2017-0154091 A1 US 2017-0024453 A1 JP 2016-219859 A

 Load balancing on distributed datastore in opendaylight SDN controller cluster (http://ieeexplore.ieee.org/document/8004238/)

The technical problem to be achieved by the present invention is to provide a shard leader selection method considering the available network bandwidth between the controllers and the available network bandwidth requirements of the shard for fast controller synchronization in a software defined network.

In the software defined network according to an embodiment of the present invention, the method of selecting a shard reader is performed by a leader selection module included in a software defined network system including a plurality of controllers.

, 및 주기 T를 입력으로 하여, 상기 주기 T마다 모든 컨트롤러들에 대한 가용 동기화 대역폭, 모든 샤드들에 대한 평균 요구 동기화 대역폭, 및 모든 컨트롤러들에 대한 한계 동기화 대역폭을 계산하는 단계; 상기 한계 동기화 대역폭의 값이 음수인 컨트롤러가 존재하는지 확인하는 단계; 상기 한계 동기화 대역폭의 값이 음수인 컨트롤러(

)가 존재하는 경우, 해당 컨트롤러(

)가 리더로 동작하는 샤드들 중에서, 평균 요구 동기화 대역폭(RSBW)이 최소인 샤드(

)를 선택하고, 해당 샤드(

)의 새로운 리더로 선정되어도 한계 동기화 대역폭(MSBW)이 음수가 되지 않는 컨트롤러(

)를 선택하는 단계; 및 상기 샤드(

)의 리더쉽을 새로운 리더로 선택된 컨트롤러(

)로 옮기고, 리더 매핑 상태

를 업데이트 하는 단계를 포함한다.
Controller Set C, Shard Set P, Reader Mapping Status

And, inputting period T, calculating the available synchronization bandwidth for all controllers, the average required synchronization bandwidth for all shards, and the limit synchronization bandwidth for all controllers, every period T; Checking whether a controller having a negative value of the limit synchronization bandwidth exists; A controller having a negative value of the threshold synchronization bandwidth

) If present, the controller (

Among the shards that act as readers, the shards with the lowest average required synchronization bandwidth (RSBW) (

), The corresponding shard (

Controller with non-negative limit synchronization bandwidth (MSBW)

Selecting); And the shard (

Controller selected as the new leader

), Leader mapping status

It includes the step of updating.

According to the leader selection method according to an embodiment of the present invention, by placing the shard leaders in the controllers in consideration of the network bandwidth, the synchronization bandwidth of the shards may not be limited by the network. There is.

According to the reader selection method according to an embodiment of the present invention, fast controller synchronization is possible in a software defined network.

The detailed description of each drawing is provided in order to provide a thorough understanding of the drawings cited in the detailed description of the invention.
1 is an exemplary diagram illustrating synchronization between SDN controllers via Raft consensus algorithm in a WAN environment.
2 is an exemplary diagram illustrating a distributed SDN structure based on the Raft consensus algorithm.
3 illustrates a shard leader selection process through a leader selection module in a distributed SDN environment.
4 shows an example of calculation of available synchronization bandwidth for each controller according to available network bandwidth between controllers.
5 is a flowchart for describing a leader selection method performed by the leader selection module.
6 shows an example of the operation of leader selection performed by the leader selection module.

Specific structural or functional descriptions of the embodiments according to the inventive concept disclosed herein are provided only for the purpose of describing the embodiments according to the inventive concept. It may be embodied in various forms and is not limited to the embodiments described herein.

Embodiments according to the inventive concept may be variously modified and have various forms, so embodiments are illustrated in the drawings and described in detail herein. However, this is not intended to limit the embodiments in accordance with the concept of the invention to the specific forms disclosed, and includes all changes, equivalents, or substitutes included in the spirit and scope of the present invention.

Terms such as first or second may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another, for example without departing from the scope of the rights according to the inventive concept, and the first component may be called a second component and similarly the second component. The component may also be referred to as a first component.

When a component is said to be "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may exist in the middle. Should be. On the other hand, when a component is said to be "directly connected" or "directly connected" to another component, it should be understood that no other component exists in the middle. Other expressions describing the relationship between components, such as "between" and "immediately between" or "neighboring to" and "directly neighboring", should be interpreted as well.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, terms such as "comprise" or "have" are intended to indicate that there is a feature, number, step, action, component, part, or combination thereof described herein, but one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, parts, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art, and are not construed in ideal or excessively formal meanings unless expressly defined herein. Do not.

First, prior to the description of the present invention, a description of the background technology of the present invention is as follows.

In Software Defined Networking (SDN), the network control and data delivery functions are separated from each other, and a logically centralized SDN controller controls the entire network equipment (eg, network switches, routers, etc.). Recently, in order to overcome the limitation of scalability of a single SDN controller, a distributed SDN controller structure including a plurality of replicated controllers has been proposed. At this time, a Raft Consensus Algorithm may be used to maintain consistency among the controllers.

In a distributed system, multiple nodes (eg servers) form groups to provide services to external clients. At this time, all nodes should maintain the same state in order to provide services even in the event of a node failure. For this purpose, when changing the state of one node, the change should be reflected in the state of other nodes. That is, nodes must agree on the changed state. In this case, the consensus algorithm is an algorithm for achieving consensus on the state of nodes in a distributed system.

The Raft consensus algorithm assumes a replicated state machine (RSM) structure. The RSM architecture assumes a distributed system consisting of multiple servers replicated. The external client sends commands to the distributed system, which then processes it and notifies the client of the completion. In the RSM architecture, it is assumed that the external client does not know the distributed structure of RSM and only considers the situation of sending commands to one server. Specifically, in the RSM structure, each server is composed of a consensus module, a log, and a state machine. The Consensus Module stores the commands received from the client in order in the local log. The Consensus Module first replicates the local log to the logs of the external servers, and then delivers it to the local state machine in the order stored in the log for processing in that order. Here, if the commands in the log are processed by the state machine, the log is said to be committed. The reason for duplicating first before committing the log is to make sure that all state machines maintain the same log and process the same commands in the same order. Here, the algorithm for guaranteeing consistency between logs is called a consensus algorithm.

The log replication process is as follows. The Consensus Module first communicates with the Consensus Modules of external servers before the commands in the local log are processed by the local state machine and replicates the local log to the logs of the external servers. When receiving the response that log replication is completed from Consensus Modules of external servers, the Consensus Module processes the commands stored in the local log in order to the local state machine and informs the client when the processing is completed. At the same time, the Consensus Module sends a message to the Consensus Modules of external servers to process commands that have been replicated but not yet processed. External servers receive the message and process the commands through their state machines.

In Raft, a node (a server in the RSM structure description) has one of three states: Leader, Follower, and Candidate. All nodes participating in the Raft group begin with a Follower state. At this time, a Raft group composed of multiple nodes can have only one Leader, and commands to all nodes in the Raft group can be processed only through the Leader. Therefore, if a leader node fails, the entire Raft group becomes out of service. To cope with this situation, the Leader node sends periodic Heartbeat messages to the Follower nodes to indicate its survival. Follower nodes maintain a Randomized Timeout value that is initialized each time they receive a heartbeat. When a timeout occurs for receiving a heartbeat message from a leader, the follower determines that the leader has failed, switches its status to Candidate, and sends a request request message to other nodes. Candidates with a majority of the votes become leaders.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

2 illustrates an example of a distributed software defined network (SDN) controller structure.

As used herein, the term "controller" refers to a functional entity that controls the relevant components (e.g., switches, routers, etc.) for controlling the flow of a flow (traffic or packet), and is a physical implementation. It is not limited to a form, an implementation position, etc. For example, the controller may be a controller functional element defined by the Open Networking Foundation (ONF), the Internet Engineering Task Force (IETF), the European Telecommunications Standards Institute (ETSI) and / or the International Telecommunication Union-Telecommunication Standardization Sector (ITU-T). entity).

The controller's Network State Database (NSDB) and applications (ie network services) are replicated across multiple controllers to support high availability and high scalability. The NSDB may be divided into shards, which are logical data partitions. In FIG. 2, the NSDB is divided into 3-shards of P1, P2, and P3.

On the other hand, shards are replicated to all controllers. That is, every controller maintains a shard-specific replica. If a replica of shard x is located on controller y, the replica is represented by Px.y, and in Figure 1 we see that controller 2 has P1.2 (that is, replica of shard 1 located on controller 2). Can be. At this time, Raft consensus algorithm is used to maintain data consistency between replicas.

3 illustrates a leader selection process through a leader selection module in a distributed SDN environment. All controllers provide Make-Leader-Local RPC, a feature that allows internal or external clients and applications to select leaders of replicas located in the controllers. The Make-Leader-Local RPC in any controller receives the Shard ID as an input and converts the replica in the controller for the shard to the Leader. For example, in FIG. 1, a Make-Leader-Local RPC (Shard 2) is transmitted to the controller 2 to move the leadership for shard 2 from P2.1 to P2.2. The leader selection module according to an embodiment of the present invention selects the leadership of the controllers through Make-Leader-Local RPC.

The leader selection module according to the present invention may be included in at least one controller among a plurality of controllers included in the SDN system, and the leader selection result is shared by the plurality of controllers.

According to an embodiment, the leader selection module may be included in an upper controller (or super controller) among a plurality of controllers included in the SDN system. That is, after leader selection is performed by the leader selection module included in the upper controller included in the SDN system having a hierarchical structure, the execution result may be provided to each of the lower controllers by the upper controller.

As used herein, the term 'module' may refer to a functional and structural combination of hardware for performing the technical idea of the present invention and software for driving the hardware. For example, the '-module' may mean a logical unit of a predetermined code and a hardware resource for performing the predetermined code, and does not necessarily mean a physically connected code or a kind of hardware.

On the other hand, in the case of a shard having a very frequent data access request, synchronization also occurs frequently, so that the leader of the shard is transferred to other controllers so that the synchronization bandwidth is not limited by the network. The controller should be located on a controller with sufficient network bandwidth. To this end, the leader selection module periodically determines the available synchronization bandwidth (ASBW) for each controller and the required synchronization bandwidth (RSBW) according to the average data access request for each shard. It is measured by the method, and the leader is selected based on this.

로 나타내고, 이는 해당 컨트롤러에서 다른 컨트롤러들로 단위시간당 전송할 수 있는 동기화 메시지들의 최대 양을 의미한다. Raft 컨센서스 알고리즘을 사용하는 임의의 컨트롤러 c에 대해,

는 해당 컨트롤러에서 다른 컨트롤러들로의 가용 네트워크 대역폭들 값들 중, 높은 상위 절반의 값을 갖는 컨트롤러들 중에 가장 작은 가용 네트워크 대역폭 값으로 결정된다. 예를 들어, 5개의 컨트롤러들 중 하나의 컨트롤러 c1에서 다른 컨트롤러들 c2, c3 ,c4, c5로의 가용 네트워크 대역폭이 각각 40, 20, 50, 10 이라고 하면,

은 가용 네트워크 대역폭이 높은 상위 2개의 컨트롤러인 c2, c4들 중 가장 작은 가용 네트워크 대역폭값을 갖는 c2의 가용 네트워크 대역폭인 40으로 결정된다. Specifically, when C is called a controller set, the available synchronization bandwidth (ASBW) for any controller c∈C is

This represents the maximum amount of synchronization messages that can be transmitted per unit time from the controller to other controllers. For any controller c using the Raft consensus algorithm,

Is determined as the smallest available network bandwidth value among the controllers with the highest upper half of the available network bandwidth values from that controller to other controllers. For example, if the available network bandwidth from one controller c1 of five controllers to the other controllers c2, c3, c4, c5 is 40, 20, 50, 10, respectively,

Is determined to be 40, the available network bandwidth of c2 having the smallest available network bandwidth value among the top two controllers c2 and c4 having high available network bandwidth.

로 나타내고, 이는 해당 샤드로의 평균적인 데이터 접근에 의해 발생하는 단위 시간당 평균 동기화 메시지들의 양을 의미한다. 즉, 데이터 접근이 잦은 샤드일수록, 해당 샤드의

는 큰 값을 가지게 된다.When P is called a set of shards, the average required synchronization bandwidth (RSBW) for any shard p 드 P is

This represents the average amount of synchronization messages per unit time generated by the average data access to the shard. In other words, the shard with more data access, the shard

Has a large value.

) 중, 자신이 Leader로 동작하는 샤드들의 동기화 메시지들을 전송하는데 사용되는 대역폭을 제외한 나머지 대역폭을 한계 동기화 대역폭(Marginal Synchronization Bandwidth ; MSBW)로 정의한다. 임의의 컨트롤러 c에 대한 한계 동기화 대역폭(MSBW)은

로 나타내고, 가용 동기화 대역폭(

)에서 컨트롤러 c가 Leader로써 동작하는 모든 샤드들에 대한 평균 요구 동기화 대역폭(RSBW)의 합을 뺀 값으로 아래와 같이 계산한다.On the other hand, any controller c has its own available synchronization bandwidth (

) Is defined as Marginal Synchronization Bandwidth (MSBW), except for the bandwidth used to transmit synchronization messages of shards acting as leaders. The limit synchronization bandwidth (MSBW) for any controller c is

, The available sync bandwidth (

) Is calculated by subtracting the sum of the average required synchronization bandwidths (RSBW) for all shards in which controller c operates as a leader.

는 샤드(Shard) p의 Leader와 컨트롤러 c간의 매핑 상태(위치)를 나타낸다. 즉, 임의의 샤드(Shard) p∈P 및 컨트롤러 c∈C에 대해,

은 샤드(Shard) p의 Leader가 컨트롤러 c에 위치한 것을 나타내고,

는 샤드(Shard) p의 Leader가 컨트롤러 c에 위치하지 않은 것을 나타낸다. here,

Represents the mapping state (location) between the leader of the shard p and the controller c. That is, for any shard p∈P and controller c∈C,

Indicates that the leader of shard p is located in the controller c,

Indicates that the leader of shard p is not located in the controller c.

)이 0보다 음수가 되지 않도록 Leader를 선정하는 것이다. 즉, 리더 선정 모듈은 모든 컨트롤러들에 대해 컨트롤러의 한계 동기화 ㄷ대역폭(MSBW)이 0보다 작은 컨트롤러가 있는지 주기적으로 확인하고, 그러한 컨트롤러가 있을 경우, Leader 재선정을 통해 모든 컨트롤러들의 한계 동기화 대역폭(MSBW)이 0 이상이 되도록 한다.
The purpose of the Leader Selection Module is to provide a marginal synchronization bandwidth for all controllers.

Leader is selected so that) is not negative than zero. That is, the leader selection module periodically checks for all controllers whether there is a controller whose limit synchronization bandwidth (MSBW) is less than 0, and if there is such a controller, the leader reselection module selects the limit synchronization bandwidth of all the controllers. MSBW) should be greater than or equal to zero.

5 is a flowchart illustrating a leader selection method performed by a controller including a leader selection module or a leader selection module according to an embodiment of the present invention.

, 그리고 컨트롤러들의 가용 동기화 대역폭(ASBW)/평균 요구 동기화 대역폭(RSBW)/한계 동기화 대역폭(MSBW)을 측정하는 주기 T를 입력받는다(S100). First, controller set C, shard set P, and leader mapping state.

And a period T for measuring the available synchronization bandwidth (ASBW) / average required synchronization bandwidth (RSBW) / limit synchronization bandwidth (MSBW) of the controllers (S100).

및

를 각각 모든 컨트롤러들 및 샤드(Shard)들에 대해 계산한다. 또한, 모든 컨트롤러들에 대해

를 계산한다(S200).Next, every cycle T

And

Is calculated for all controllers and shards, respectively. Also, for all controllers

To calculate (S200).

의 값이 음수인 컨트롤러 c가 존재하는지 확인한다(S300).

의 값이 음수인 컨트롤러 c가 존재하지 않을 경우, T 시간이 지난 후 다시

의 값이 음수인 컨트롤러 c가 존재하는지 확인한다. to the next

Check whether the controller c having a negative value is present (S300).

If no controller c with a negative value of c does not exist, retry after T time

Check if there is a controller c with a negative value.

의 값이 음수인 컨트롤러 c가 존재할 경우, 해당 컨트롤러(즉,

)가 Leader로 동작하는 샤드(Shard)들 중, RSBW가가 최소인 샤드(즉,

)를 선택하고, 해당 샤드의 새로운 Leader로 선정되어도 한계 동기화 대역폭(MSBW)이 음수가 되지 않는 컨트롤러(즉,

)를 선택한다(S400). if

If a controller c with a negative value of c exists, that controller (that is,

Among the shards where) acts as the leader, the shards with the smallest RSBW price (ie

), The controller whose non-negative limit bandwidth (MSBW) is not negative even if it is selected as the new leader of the shard (i.e.

Select (S400).

에게 Make-Leader-Local RPC(

)를 전송하여

의 Leadership을

에서

로 옮기고, 이에 맞게

를 업데이트 한다(S500).Next, the leader selection module

Make-Leader-Local RPC (

) By sending

Leadership

in

Move to

Update the (S500).

) 값이 음수인 컨트롤러가 존재하지 않을 때까지 반복한다.
This process of moving leadership is not limited to limit synchronization bandwidth (

Repeat until no controller with negative value exists.

6 illustrates an example of an operation of a leader selection method performed by a leader selection module (or a controller including a leader selection module). Assume a distributed SDN environment consisting of three controllers C1, C2, C3 and three Shards P1, P2, and P3. The left-hand table based on the dashed lines shows the available synchronization bandwidth (ASBW), the average required synchronization bandwidth (RSBW), and the limit synchronization bandwidth (MSBW) values of the initial state. Referring to FIG. 6, it can be seen that the limit synchronization bandwidth (MSBW) of C2 is negative. This is because the available synchronization bandwidth (ASBW) of C2 is 15, but since the sum of the average required synchronization bandwidth (RSBW) of the shard leaders (i.e., P2) located at C2 is 30, the limit synchronization bandwidth (MSBW) of C2 is -15. To be. If this situation persists, the synchronization for P2 may be delayed, resulting in a significant degradation in data access performance for P2.

In the method of selecting a shard reader according to an embodiment of the present invention, a controller C2 having a negative limit synchronization bandwidth (MSBW) is identified, and a shard having the smallest average required synchronization bandwidth (RSBW) among shard leaders located at C2. Find it. Meanwhile, in FIG. 6, since there is only one shard leader located in C2 (porridge, leader for P2 shard), the leadership of the shard is moved from C2 to another controller. At this time, C3, a controller having a limit synchronization bandwidth (MSBW) greater than 30, which is the average required synchronization bandwidth (RSBW) of P2, is selected, and the reader selection module sends Make-Leader-Local RPC (P2) to C3 to C3. Have P2 leadership. The table on the right of the dotted line shows the state after P2's leadership has moved from C2 to C3, and it can be seen that the limit synchronization bandwidths (MSBW) of all controllers are positive.

Although the present invention has been described with reference to the embodiments illustrated in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and equivalent other embodiments are possible. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

100: controller 110: shard
200: reader selection module

Claims (5)

A method of selecting a shard leader performed by a leader selection module included in a software defined network system including a plurality of controllers, the method comprising:
(a) Controller set C, shard set P, leader mapping status

And a period T for measuring the available synchronization bandwidth (ASBW) / average required synchronization bandwidth (RSBW) / limit synchronization bandwidth (MSBW) of the controllers as inputs, and the available synchronization bandwidth for all controllers at each period T (

), Average required sync bandwidth for all shards (

), And the limit synchronization bandwidth for all controllers (

Calculating;
(b) the limit synchronization bandwidth (

Checking whether there is a controller having a negative value;
(c) the limit synchronization bandwidth (

Controller with negative value ()

) If present, the controller (

Among the shards that act as readers, the shards with the lowest average required synchronization bandwidth (RSBW) (

), The corresponding shard (

Controller with non-negative limit synchronization bandwidth (MSBW)

Selecting); And
(d) the shard with the minimum required synchronization bandwidth (RSBW) (

Leadership of the limit synchronization bandwidth (

Controller with negative value ()

Controller selected as the new leader ()

), Leader mapping status

Updating the;
Step (d) is the limit synchronization bandwidth (

Is performed repeatedly until no controller with a negative value of) exists,
How to choose a shard leader.
The method of claim 1,
The limit synchronization bandwidth (

If no controller with a negative value is present, the limit synchronization bandwidth (

Checking whether a controller having a negative value of) exists.
How to choose a shard leader.
delete A method of selecting a shard leader performed by a leader selection module included in a software defined network system including a plurality of controllers, the method comprising:
(a) Controller set C, shard set P, leader mapping status

And a period T for measuring the available synchronization bandwidth (ASBW) / average required synchronization bandwidth (RSBW) / limit synchronization bandwidth (MSBW) of the controllers as inputs, and the available synchronization bandwidth for all controllers at each period T (

), Average required sync bandwidth for all shards (

), And the limit synchronization bandwidth for all controllers (

Calculating;
(b) the limit synchronization bandwidth (

Checking whether there is a controller having a negative value;
(c) the limit synchronization bandwidth (

Controller with negative value ()

) If present, the controller (

Among the shards that act as readers, the shards with the lowest average required synchronization bandwidth (RSBW) (

), The corresponding shard (

Controller with non-negative limit synchronization bandwidth (MSBW)

Selecting); And
(d) the shard with the minimum required synchronization bandwidth (RSBW) (

Leadership of the limit synchronization bandwidth (

Controller with negative value ()

Controller selected as the new leader ()

), Leader mapping status

Updating the;
The limit synchronization bandwidth (

) Is the available synchronization bandwidth (c) of any one of the plurality of controllers (c)

) Is defined by the equation as the remaining bandwidth except the bandwidth used to transmit the synchronization messages of the shards that the controller (c) acts as a reader,
The equation is

ego,
remind

Is a binary variable representing the mapping state of the reader and controller (c∈C) of any shard (p∈P),

Means that the leader of the shard (p) is located in the controller (c),

Means that the leader of shard p is not located in controller c,
How to choose a shard leader.
A method of selecting a shard leader performed by a leader selection module included in a software defined network system including a plurality of controllers, the method comprising:
(a) Controller set C, shard set P, leader mapping status

And a period T for measuring the available synchronization bandwidth (ASBW) / average required synchronization bandwidth (RSBW) / limit synchronization bandwidth (MSBW) of the controllers as inputs, and the available synchronization bandwidth for all controllers at each period T (

), Average required sync bandwidth for all shards (

), And the limit synchronization bandwidth for all controllers (

Calculating;
(b) the limit synchronization bandwidth (

Checking whether there is a controller having a negative value;
(c) the limit synchronization bandwidth (

Controller with negative value ()

) If present, the controller (

Among the shards that act as readers, the shards with the lowest average required synchronization bandwidth (RSBW) (

), The corresponding shard (

Controller with non-negative limit synchronization bandwidth (MSBW)

Selecting); And
(d) the shard with the minimum required synchronization bandwidth (RSBW) (

Leadership of the limit synchronization bandwidth (

Controller with negative value ()

Controller selected as the new leader ()

), Leader mapping status

Updating the;
The controller selected as the new leader (

The step of transferring the leadership to the controller selected as the new leader (

) Uses the Make-Leader-Local RPC to select which controller (

Transferred to)
How to choose a shard leader.

Images