CN1284094C - Distributed parallel scheduling wide band network server system - Google Patents

Abstract

The present invention discloses a distributed parallel scheduling network server system. A client terminal and a server terminal are matched. Scheduling message pick-up and scheduling request processing are asynchronously carried out. Each node in a server group can accept and process a scheduling request. A scheduling process embodies distributed parallel. An input output scheduling interface in the mode of distributed parallel is realized, and the scheduling bottleneck problem existing in the prior art is thoroughly eliminated. When the present invention is used, more request users can be supported on the basis of the scale of the existing hardware system. The expandability of the system and the load capacity for a user can be greatly enhanced, and therefore, the integral price performance ratio of the server system is raised.

Description

A kind of method that makes up the distributed parallel scheduling network server system

Technical field:

The present invention relates to the webserver, particularly wide band network server, as: wideband video server, broadband data services device etc.

Background technology:

The mode that existing wide band network server adopts has two kinds, with the traditional server of single large scale computer formation and the group system server of forming with a plurality of power PC machines.Traditional server not only costs an arm and a leg, and under the continuous condition of enlarged of client access amount, the reliability deficiency, calculating, processing, storage capacity and system bandwidth are difficult to expansion, cause congested situation easily.Computer cluster allows the accurate personal computer control of thousands of station symbols lot of data intensive task, the computing cost is lower than the computing cost of present large computer system, and has characteristics such as high availability, high system throughput, high expandability, low cost, low user's operating lag.

Group system CLUSTER: multiple servers (high-performance PC) connects to form with the High Speed General network interconnection device, and the operation relative program, uniform dispatching, and Coordination Treatment realizes the system that efficient parallel is handled.Have and to bear a plurality of users that try one's best/processing (request) task, high availability, characteristics such as high expandability simultaneously.

Existing wide band network server situation:

1.TurboCluSter

TurboCluster is the cluster scheme of an enterprise-level, and it allows to make up high available, extendible network on a plurality of computing machines.Use the TurboCluster system can improve the service quality of serving based on the multiple network of ICP/IP protocol significantly, these services comprise Web, Mail, News and Ftp etc.TurboCluster has good usability, extensibility and manageability, and the quantity of the real server of cluster inside can expand, and its structure as shown in Figure 1.

When client computer when server cluster system sends a request, this request at first arrives the Advanced Traffic Management device, the Advanced Traffic Management device is forwarded to this request on a certain the real server in the cluster by certain scheduling strategy and handles, and final response request will directly send to client computer.Because final response request not by senior browsing management device but directly send to the client, has alleviated the load on the Advanced Traffic Management device so greatly, thereby has reduced the possibility of bottleneck generation.The scheduling strategy that adopts among the TurboCluster has: samsara (Round Robin), weighting samsara (Weighted Round Robin), minimum connection (Least Connection).Produce the possibility that lost efficacy in order to reduce the Advanced Traffic Management device, TurboCluster has prepared a backup machine for the Advanced Traffic Management device.This backup machine inquires that constantly manager confirms it just in operate as normal, in case find that main manager lost efficacy, backup machine will be taken over it and work on.

2.LinuxVirtualServer

(Linux Virtual Server LVS) builds on the server cluster of reality the Linux virtual server, and the user can't see many real server that service is provided, and can only see a server as load balancer.Actual server connects by high-speed local area network.The front end of real server is a load balancer, and he finishes user's request scheduling on real server, seems all services so all to finish by virtual server.The Linux virtual server can provide good extensibility, reliabilty and availability.The user can increase or reduce a node pellucidly, can monitor real server, has node failure just to reconfigure system if find.The structure of Linux Virtual Server as shown in Figure 2.The client uses VIP (virtual IP address) address access server.The scheme of its load balance is:

1) network address translation (nat) technology: load balancer is after receiving services request (IP bag), press load balancing algorithm, determine real server, this link information is write IP request connection table, the destination address of this IP bag and port numbers are changed into this server by the VIP address IP address and port numbers simultaneously, transmit this IP bag, real server is after finishing the request task, send load balancer back to replying the IP bag, load balancer changes the source address of replying the IP bag VIP address into and is forwarded to external network again.

2) IP tunnel (IP-TUNNEL) technology: load balancer is behind the services request IP that receives from external network, the encapsulation source IP address, then with data packet form, be forwarded to a certain real server, real server is after finishing the request task, directly by source IP address the result is returned the client computer of sending services request, the load balancer of this method is no longer born and is transmitted the task of returning the IP bag, and load has just been lacked than the NAT method.

3) direct route (DIRECT ROUTING) technology: the LVS of this method is different, load balancer and real server all are in same section LAN (Local Area Network), load balancer receives the services request IP bag of client computer, load balancer is distributed to a certain real server, directly revise the MAC Address of the MAC Address of link layer then for this real server, to this real server, after finishing request, real server also is directly the result to be delivered to client computer the IP packet transfer.This strategy has further reduced the load of load balancer self.

From above system as can be seen, in order to maximally utilise all resources in the cluster, group system need have the balancing dynamic load function, and it monitors the loading condition of the actual node in the cluster and the change of dynamically dispatching.We can say that the quality of scheduling feature directly influences the integrity service ability of system.No matter be the scheduling mode that adopts the Advanced Traffic Management device among the TurboCluster, or the scheduling scheme of realizing in the LVS virtual server based on NAT, IP-TUNNEL or DIRECT ROUTING technology, all can abstractly be the centralized scheduling model of Fig. 3.

Client requests is received and queuing by single dispatch interface, is transmitted to actual server one by one according to dispatching algorithm.Obviously along with the expansion of system scale (for the advanced direct routing mode of technology, maximum service device node number is above 100) and the increase of user capture amount, dispatch interface will become system bottleneck.This makes when having a large amount of client requests to arrive, just can be because of the serious overload of dispatch interface main frame, and make response time may become very long to client requests, surpass client's degrees of tolerance, even the system that can think by mistake is unavailable.More seriously, if this special main frame collapse, total system is promptly unavailable, even select another again automatically or start guest machine as new dispatch interface main frame, does not solve but bottleneck problem is still basic.

Summary of the invention

The technology of the present invention purpose is to realize a kind of wide band network server system of distributed parallel scheduling interface, solves the bottleneck problem of existing group of planes formula broadband server to the client requests scheduling, shortens the response time of client requests, the availability of enhanced system; Greatly improve the extensibility and the load capacity of system, thereby improve the cost performance of server system integral body.In order to reach this purpose, the technical solution used in the present invention is:

Client computer is cooperated with server zone, and schedule information is picked up with dispatch request and handled asynchronous carrying out; In the server zone each node can both accept, processing and response scheduling request, scheduling process is presented as distributed parallel.The system flow block diagram of client requests scheduling as shown in Figure 4, its scheduling model is as shown in Figure 5.

By realizing the input/output scheduling interface of distributed parallel of the present invention, can thoroughly eliminate the scheduling bottleneck problem that exists in the prior art.

Job step of the present invention is:

Enforcement of the present invention is divided into asynchronous two processes (as Fig. 4): server info picks up and maintenance process B and client requests scheduling process A.

In the period of not initiating client requests as yet, operate in each the server node information B1 of dispatch service end program interaction in each server, the dispatching client program that operates in the client computer is communicated by letter with the dispatch service end program timing in the server, each active server node and information B2 such as COS that provides and service ability thereof in the nearest period server zone are provided, and in this machine, safeguard the relevant information table.

When the client access server, the access interface that the client requests that client applications in the client computer sends provides by the dispatching client program obtains an effectively address A1 of (movable and have related service) server at random from the information table of this machine, and initiates request A2 to this server.Owing to be the server address of obtaining at random, the request of different client computers can be distributed to different server nodes randomly, and this has just solved all requests enter server system from same dispatch interface bottleneck problem;

Then, the node server that receives client requests returns to real server address of client according to current system status and dispatching algorithm, because current server is also at optional row, usually preferentially chosen, this moment, client computer will directly establish a communications link with it, begin to provide the application service of customer requirement, finishing scheduling;

If this server load overweight (by the dispatching algorithm decision), can not provide service to this request, the meeting that then returns to the client is the address of the lighter server of load, client requests need not postpone to wait for, continuation sends request to new server, thereby can be accepted and provide service by this server.

The worst situation is, the entire system load is all very high, client requests will attempt connecting (connect time delay at every turn and be about more than ten millisecond) one by one according to the server address that dispatch service end program is returned this moment, have only and when Servers-all all can not be served, just understand discard request, though can not occur that the sort of server still has service ability in the aforementioned group system, but because of request message blocks at dispatch interface, and make the unaccepted situation of request, improved the availability of system.

Description of drawings:

Fig. 1 is turbocluster server architecture figure

Fig. 2 is Linux virtual server architecture figure

Fig. 3 is the centralized scheduling illustraton of model

Fig. 4 is the system flow block diagram

Fig. 5 is the distributed parallel scheduling illustraton of model

Fig. 6 is a VOD service system structural drawing

Wherein, c1, c2, c3...: the client computer of visit book server, be general purpose personal computer PC, be connected into Internet by 100Mbps Ethernet adapter; Installation and operation windows2000 operating system, VOD client's application software and client software module of the present invention.

LAN (Local Area Network): the physical network path of inner each node of Connection Service device group; Adopt 100Mbps Ethernet high speed InterWorking Equipment; Internet: the network path that connects client computer and server;

S1, s2, s3, s4...: general high-performance PC (band 100Mbps Ethernet adapter) provides VOD service as server; Installation and operation linux operating system; Installation and operation service end software module of the present invention and VOD service application software.

Fig. 7 is the dispatcher software structural drawing that is applied to the VOD server

Wherein, dispatch service end program and dispatching client program all adopt C Plus Plus to write.

Fig. 8 is a node administration modular program process flow diagram

Fig. 9 is application server interface modular structure figure

Figure 10 dispatches the primary module structural drawing

Figure 11 is scheduling sublayer modular structure figure

Embodiment:

For the embodiment that reaches technique effect uses for VOD (video on demand) video on-demand system.Below in conjunction with accompanying drawing it is elaborated.

Video on-demand system: pass through in the high speed Internet access technical battery various wideband multimedias website and carry out various interacting activities such as on-line film, TV, online music, long-distance education, meeting, medical treatment.

Implement to divide two stages:

Phase one: client computer obtains server zone up-to-date information, the A process of corresponding diagram 4.

Step 1: start dispatch service end program

Require in the server system every main frame (s1, s2, s3, s4) traffic control serve end programs all.Dispatch service end program is the finger daemon of multithreading form, and as Fig. 7, this program comprises node administration, task scheduling, and application service interface and load pick up and safeguard etc. submodule.The program maintenance of dispatch service end has four tables:

Service name-port numbers mapping table---in the table each is the mapping of a service name-port numbers, is used to describe the access entrance of application service.

Main frame upgrades concordance list---determine to send infomational message to any platform node main frame when being used for more the new node host information.

The load on host computers table---write down the loading condition of all mobile host computers in the system.

Service-main frame mapping table---write down the information of the respective services that each service host provides.

Start dispatch service end program and will at first cause the operation of node administration submodule.Its flow process such as Fig. 8 are described below:

1, carries out netinit work: create TCP communication socket, UDP communication socket and be tied to the port of appointment, create the UNIX communication socket.

2, with program initialization be demons, the realization in 1,2 steps is medium programmer to be familiar with, and does not describe in detail at this.

3, this nodal information of broadcasting in server zone states that to other mobile host computers nodes this machine adds system with this.

4, produce a thread, regularly check and processing host is upgraded each host information in the concordance list: it is that ascending order with the markers RecvTime that receives this record sorts that main frame upgrades concordance list, handles and carries out from the beginning to the end.If the difference of the RecvTime of first element and current time is less than a update cycle UPD_HOST_TIME (defining with grand mode according to system scale), then do not do any operation, wait for next time and upgrading, because the All hosts that main frame upgrades in the concordance list has all sent message to this machine in a update cycle.The deletion main frame upgrades the dead main frame in the concordance list, and the difference of RecvTime and current time is all thought dead main frame greater than the element of three update cycles.Behind the dead main frame of deletion, need the respective host record in deletion load on host computers table, the service-main frame mapping table.In addition, be that these dead main frames are exclusive if some service is arranged, also in service name-port numbers mapping table, delete corresponding information on services.At this moment, it is that these main frames just need upgrade between a update cycle and three update cycles that will there be the RecvTime of some elements and the difference of current time in the foremost that main frame upgrades concordance list---send this host information message by the UDP communication socket to them.But to these main frames is not update all, but therefrom selects a part to upgrade, and the maximum number of renewal is 1/3 of all mobile host computers numbers, and the design of this number is to consider to upgrade in three update cycles to finish all mobile host computers.

5, produce a thread, whether the service in quantitative check local service name-port numbers mapping table is dead.Because each element in local service name-port numbers mapping table all has a server processes attribute that service is provided, just whether activity checks whether service is dead to the state of a process by checking this process number correspondence.This thread starts a timer simultaneously, extracts this machine load information by system call interfaces, upgrades the item of this machine load in main frame-loading liquifier.

6, adopt the multiplexed select of I/O () to monitor a TCP communication port (with client communication), a UDP communication port (intercommunication of server node machine) and a UNIX territory socket port (communicating by letter) simultaneously with application service process.

7, main thread enters and monitors circulation, waits for receiving each port message and producing each Message Processing thread respectively.

Message Processing is described:

TCP message: being initiated by the dispatching client program, mainly is the mapping of client's update service host information and name Service.

Then generate a thread process when dispatch service end program listens to a TCP message, treatment scheme is as follows:

1. read message from the TCP socket;

2. the version number in the judgement message if version number is consistent with the system information current version, then sends an empty data message, otherwise, send the service host information of redaction and serve map information (from each table, obtaining);

3. close socket, finish thread.

UDP message is the communication between the service node machine, is used for mutual node host information (load, service).

Then generate a thread when arriving a message, treatment scheme is as follows:

1. sense data is reported in the socket, and verifies its legitimacy;

2. according to the type of message, do different processing;

2.1 node machine update request kind of message

1. according to the address of information source main frame, the lastest imformation of in upgrading concordance list, searching this main frame.If current time and this main frame lastest imformation, interval＜timer time T of transmission markers send_time, then explanation has sent message to information source main frame (being equivalent to reply), upgrade the reception markers recv_time of information source main frame, and sort according to the direction that the recv_time of main frame successively decreases.Change 2..If the time interval＞T, then the information of replying by each table organization that safeguards in this node machine sends to the information source main frame, upgrades send_time and recv_time simultaneously, and sorts according to the direction that the recv_time of main frame successively decreases.Change 2..If do not find this host information, illustrate to receive the broadcast that a new server node that starts sends, also should organize the message of local host, send to the information source main frame, reply.In the renewal concordance list of this machine, increase the send_time and the recv_time of this information source main frame then, and sort according to the direction that the recv_time of main frame successively decreases.

2. according to message, update service-main frame mapping table, service name-port numbers mapping table;

3. from message, obtain load information, upgrade main frame-loading liquifier;

4. close socket, finish thread.

2.2 the node machine upgrades acknowledgement type message

1. according to the address of information source main frame, the lastest imformation of in upgrading concordance list, searching this main frame.Upgrade the recv_time of information source main frame, and sort according to the direction that the recv_time of main frame successively decreases.

2. basis, update service name-port numbers mapping table, service-main frame mapping table;

3. the load terms from message obtains load information, upgrades main frame-loading liquifier;

4. close socket, finish thread.

The UNIX domain message: communicate with the VOD application server, mainly be application server to this scheduler program registration/cancellation related service, obtaining the scheduling support in whole bundle of services, and initiate dispatch request by this message.

Then generate a thread when arriving a message, treatment scheme is as follows:

1. from socket, read message, and verify its legitimacy;

2. according to the type of message, do different processing;

1. request registration kind of message: in service name-port numbers mapping table, increase service entry, revise service-server mapping table.

2. kind of message is nullified in request: delete service entry in service name-port numbers mapping table, revise service-server mapping table.

3. request scheduling kind of message: call the scheduling sublayer module, and will return server ip address and return to application server by UNIX territory socket.

3. close socket, finish thread.

Step 2: start VOD server module program

In order to use the function of software of the present invention, the VOD server program is a kind of as broadband service, when starting, need call the interface function (providing) that this software provides with the header file form, by aforesaid UNIX domain message to its application service of this software dispatch service end program registration---Service name and access port.

Step 3: client computer (c1, c2, c3) traffic control client-side program

The dispatching client program is divided into backstage executive routine UPD.EXE and dynamic link library SchedClientPro.DLL., UPD.EXE requirement operation before client applications such as the operation of VOD CLIENT PROGRAM generally is arranged in and calls the interface function startup that dynamic link library provides by the os starting operation or by VOD CLIENT PROGRAM authorized personnel before the client opens CLIENT PROGRAM.When this program is installed, can generate the listing file of an available server IP addresses of nodes in this machine system directory, address wherein is easy to guarantee to be available by installation process.

Program UPD.EXE safeguards effective server address table, the tabulation of service name-port mapping, the tabulation of service-server mappings, its execution flow process such as Figure 10 with document form in this machine system directory.At first obtain a server address at random from effective list of server addresses, then check server info version number, this version number is kept in another file in the system directory, and the up-to-date information that is used for guarantee information and server zone is synchronous.If do not look for this file (when initial the installation), then generate it, and to put initial version number be 0; Initiate TCP to the system call port of the server that obtains and connect, if other address is then attempted in connection failure, all unavailable just failure is returned until all addresses; As described above, connect by this, every up-to-date information of dispatch service end program organization server zone returns to this machine, and revises each table in this locality, successfully returns.After this, the dispatching client program can regularly repeat above operation, and when guaranteeing that client applications is initiated request, total energy obtains up-to-date effective service device address.

Step 4: start the VOD client-side program

The mode common according to the application software of windows operating system starts, and present embodiment does not have specific (special) requirements.

Subordinate phase: server scheduling client requests, the B process of corresponding diagram 4.

Step 5:VOD CLIENT PROGRAM is called the interface function of the dynamic link library mode that this software provides, and obtains one or more effective server addresss at random from the scheduling CLIENT PROGRAM.

Step 6:VOD CLIENT PROGRAM is initiated the TCP connection request according to the application schedules port of the VOD service routine of related application agreement on this server node machine of VOD service oneself, successful connection then sends the application schedules request message, when few situation connection is unsuccessful, think that then this server address is invalid, choose another server address and initiate connection request.After the VOD service routine accepts request and connects, send the request scheduling kind of message by UNIX cover joint word to aforesaid node administration submodule again.

Step 7: as Fig. 8, the node administration submodule is received a UNIX message, will produce a sub-thread, by its calling task scheduling sublayer module interface function, and requires to return a scheduling result---the server node IP address of application service is provided.Return with giving the VOD service routine by former UNIX socket then.As Figure 11, the task scheduling submodule is embodied as a plurality of extendible dispatching algorithms, as take turns commentaries on classics service method, weighted round robin service method, minimum connection elder generation service method, local first method, dispatching algorithm such as copy Priority Service method is arranged, non-the present invention of specific implementation of algorithms of different limits, and does not give unnecessary details at this.Can import different parameters according to the different call functions of application type, make the different application server can choose only dispatching algorithm, can adopt the first service method of minimum connection as http server, ftp server can adopt minimum bandwidth use amount precedence method.Present embodiment takes minimum linking number to add this machine Priority Service algorithm.

Step 8:VOD service routine communicates to connect by TCP and returns this scheduling result to the VOD CLIENT PROGRAM; The VOD CLIENT PROGRAM begins therewith the data in server transmit port immediately to be set up the application data transmission and connects, and successful connection then can finish scheduling (it happens frequently for actual conditions).Connection failure under few situation thinks that then the service failure of this address gets back to step 5 and repeat.

From above step as can be seen: client requests mails to the arbitrarily effective server the server zone at random, can receive total identical that requested service device number will be with active server simultaneously, the bandwidth of solicited message is N times (N is the server node number) of existing system, and receive the requested service device can be according to dispatching algorithm reorientation client requests, further prevented from processing of request is blocked in the request queue of a station server, realized the function of balancing dynamic load.Thus; we may safely draw the conclusion: the present invention has thoroughly solved the problem of scheduling bottleneck between VOD client and the VOD server system; implement the present invention and can on existing hardware system scale basis, support more to put broadcasting user; greatly improve the extensibility of system and to user's load capacity, thereby improve the cost performance of server system integral body.

Embodiment of the invention technique effect compares:

An existing broadband services VOD system satisfies 6000 users' VOD program request at present, supports 300 users' normal concurrent program request simultaneously, and the time that stream connects is no more than 20 seconds.Its media server adopts Sun 6000E system, can store the above video frequency program of 300G, but system cost is more than 2,000,000 Renminbi.

And embodiment of the invention system: the number of users that can use simultaneously: the 100--10000 family, server response time adopts the present invention program's the system that reaches or better performance index identical with last plane system only to need less than 500,000 Renminbi less than 2 seconds.

Claims (2)

1. A method for building a distributed parallel scheduling network server system, realizes the following steps by server and client computer: It adopts two asynchronous processes: the server picks up, maintains information and accepts, processes and responds to the client request process (B) and the client obtains server group status information and request scheduling process (A); From the time period when the client request has not yet been initiated, the scheduling server running in each server will exchange the information of each server (B1), and the scheduling client running in the client computer and the scheduling server in the server will communicate regularly to obtain the server in the latest time period. Information (B2) on each active server node in the group and the service type and service capability it provides, and maintain an information table containing the above-mentioned obtained information (B2) in the client computer; When the client computer accesses the service of the server, the application client in the client computer requests to query the information table maintained in the client computer containing the obtained information (B2) through the access interface provided by the scheduling client computer. Randomly obtain the address of one or a group of valid servers (A1), and take a valid server to initiate a request to it (A2), where a valid server refers to an active server with related services; Then, the server that receives the request from the client computer returns a real server address to the requested client computer through a scheduling algorithm according to the current system status of each server. At this time, the client computer will directly establish an application communication connection with it, and the server begins to provide the application service required by the client computer. , the scheduling ends; If the selected server is too heavily loaded to provide services to the client, the client's request does not need to be delayed, and continues to select one of other effective servers and send a request to it, so that it can be accepted and provided by the newly selected server. Serve. 2. a kind of method of building distributed parallel scheduling network server system according to claim 1 is characterized in that the VOD video-on-demand system based on distributed parallel scheduling network server realizes the following steps: Phase 1: The client obtains the latest information about the server farm Step 1: Start the scheduling server Each server host in the server system runs a scheduling server, which is a multi-threaded daemon process, including node management, task scheduling, load picking and maintenance sub-modules, starting the scheduling server will first trigger the node management sub-module to run , and its flow is: 1. Perform network initialization: create TCP sockets, UDP sockets and UNIX sockets, and bind them to specified ports; 2. Initialize the process as a daemon process; 3. Broadcast the information of this node in the server, and declare to other active host nodes that this machine will join the system; 4. Generate a thread to regularly view and process the host to update the host information; 5. Generate a thread to regularly check whether the service in the local service name-port number mapping table is dead; the thread starts a timer and extracts the load information of the local server through the interface; 6. Use I/O multiplexing select() to simultaneously monitor a TCP communication port to communicate with the client computer, a UDP communication port to realize mutual communication between node machines, and a UNIX domain socket port to complete and apply Service process communication; 7. The main thread enters the loop, receives messages from each port and processes them; Step 2: Start the VOD server machine Step 3: The client runs the dispatch client The execution process of the dispatching client is: firstly obtain a server address randomly from the address list file, and then check the version number of the server information, which is stored in another file serverVer in the system directory to ensure the latest information, if If this file is not found during the initial installation, it will be generated and the version number will be set to 0; a TCP connection will be initiated to the system scheduling port of the obtained server. If the connection fails, another address will be tried until all addresses are unavailable. Return; through the connection, obtain the information of the server group, and generate the latest server IP address list, service-server mapping table and service name-service port mapping table on the client, and return successfully; after that, the scheduling client will repeat regularly The above operations ensure that when the client application initiates a request, it can always obtain a valid server address; Step 4: Start the VOD client Phase 2: Server dispatches client requests Step 5: the VOD client calls the interface of the dynamic link library mode, and obtains an effective server address randomly from the scheduling client; Step 6: The VOD client initiates a TCP connection request to the application scheduling port of the VOD server on the server according to the relevant protocol of the VOD service itself. If the connection is successful, the application scheduling request information is sent. This server address is invalid, return to step 5, select another server address to initiate a connection request; after the VOD server machine accepts the request connection, it sends a request scheduling type message to the aforementioned node management submodule UNIX socket; Step 7: The node management sub-module will generate a sub-thread upon receiving a UNIX message, which will call the interface function of the task scheduling sub-module, and request to return a scheduling result—server address; then return to VOD through the original UNIX socket The server machine; the task scheduling sub-module is implemented as multiple scalable scheduling algorithms, including the scheduling algorithm of the round-robin service method, the weighted round-robin service method, the least connection first service method, the local priority method, and the copy priority service method; Step 8: The VOD server returns the scheduling result to the VOD client through the TCP communication connection; the VOD client immediately starts to establish an application data transmission connection with the server, and ends the scheduling if the connection is successful.

Images