CN1276629C - Flow-media cluster service content scheduling method based on Netfilter architecture - Google Patents

Abstract

The present invention relates to a flow-media cluster service content scheduling method based on Netfilter architecture, which comprises: a, the user end sends a service request to a cluster; b, after a dispatcher receives a service request data packet, the data packet is forwarded to the corresponding server or is sent to the user space according to the condition whether an active record connected with the service request is in a dispatch table; the data packet is forwarded to the corresponding server or is sent to the user space of the server; c, a server storing the required program is found out in a program database in the user space according to the data packet content; a target server is selected by considering the geographical distribution and the loading condition of the server, and the dispatch record is added into the dispatch table; d, the dispatcher establishes TCP connection from the user end to the target server, and the data packet is forwarded; e, after the target server receives the forwarded the data packet, the TCP response connection from the target server to the user end is established, and the required program is sent to the user end. The service cluster is unitedly managed by making use of the dispatcher; the intelligence dispatch of the user request is realized, and system efficiency is high.

Description

Streaming media cluster service content dispatching method based on the Netfilter framework

Technical field

The present invention relates to the computer network, multimedia technology, specifically is a kind of streaming media cluster service content dispatching method based on the Netfilter framework.

Background technology

Stream media technology is a kind of online play-back technology of rising in recent years.Online broadcast, as the term suggests be exactly to play while downloading, player also need not wait until that multimedia file all downloads to local hard drive and plays later on again, but plays when downloading.Player hews out a buffering area in the internal memory of user side; the space of several Mbytes normally; player begins the decoding broadcast after waiting until enough data being arranged in the buffering area; this process needs the time in several seconds usually; saved the stand-by period with playing to compare greatly again after having downloaded; and data directly are stored in the internal memory, promptly be played the device deletion after playing, therefore digital publishing rights is played a protective role.Stream media technology is widely used in the information service field of the Internets such as internet multimedia news briefing, online live, the web advertisement, ecommerce, video request program, long-distance education, network radio station, video conference.At present, the MOV form of the Quicktime of the RM form of RealNetwork company, Apple company and the ASF of Microsoft use more streaming form on the Internet.

Along with the continuous development of various system applies, more and more higher to the requirement of the computing power and the network bandwidth, Clustering is just arisen at the historic moment.So-called cluster is exactly by the multicomputer system of network with one group of separate loose coupling of computer formation, is managed with the pattern of one-of-a-kind system.Load balancing cluster generally is made up of foreground scheduler and background server, and scheduler is responsible for the task distribution, and server is responsible for responding user's request.Whole group system externally provides a virtual ip address, the user needn't understand the reason distribution of the interior service of cluster implements and be connected situation, just can service request be proposed as long as know this virtual ip address to service cluster, whole group system is transparent to the user, just as same high-performance computer.The maximum characteristics of this framework can realize load balancing exactly, the task scheduling server can be current according to each background server load state, service request is distributed to suitable server, like this, the overall performance of cluster can near linear growth, also improve service quality simultaneously along with the increase of background server number.

Provide streaming media service by cluster mode, improved the overall performance of system to a great extent, but because load balancing cluster is a distributed systems, the function opposite independent of each service node, the film source of storing on the different streaming media servers may be also inconsistent, this just requires scheduler to understand sheet source distribution on each server, according to the distribution of content situation of film source data-base recording request is distributed to suitable servers.

At present, the content scheduling scheme that computer network adopts has two kinds: a kind of scheme is to share a large storage facilities by multiple servers, the film source of each station server storage is in full accord like this, only need during the dispatch service request to carry out according to the server present load, but done significant limitation like this, at first, each server and memory device will be in the same network segments, and this has just limited the flexibility that cluster distributes.Secondly, the memory device of sharing itself may become the bottleneck of whole system.

Another kind of scheme is that content distribution server returns to user side with current program profile information and load state, select service node by user side again, this realization is based on the selection of user side, the mechanism that lacks united and coordinating during scheduling, system flexibility is not high, and may have many potential safety hazards.

Netfilter (network filter) framework is the network security capability framework of Linux2.4 kernel, be positioned at the core layer of (SuSE) Linux OS, it is placed with some checkpoints (HOOK) in flow through a network, registration has different processing functions on each HOOK point, can realize Packet Filtering, processing data packets, network address translation, and multiple network safety function such as anti-attack.

Fig. 2 is a Netfilter framework schematic diagram.In procotol IPv4 (Internet Protocol Version4), defined five checkpoints (HOOK) of Netfilter framework, its title and function are as follows:

[1], NF_IP_PRE_ROUTING: the packet that has just entered network layer can carry out version number at this and check address filtering etc. by this HOOK;

[2], NF_IP_LOCAL_IN: be sent to this HOOK that passes through of this machine after route is searched, packet has just entered user's space then;

[3], NF_IP_FORWARD: the bag that transmit is by this HOOK, and packet just can directly be forwarded without user's space like this;

[4], NF_IP_LOCAL_OUT: the packet that this machine process is sent is by this HOOK, and packet changes the (SuSE) Linux OS kernel spacing over to from user's space thus;

[5], NF_IP_POST_ROUTING: all will pass through this HOOK by the bag that the network equipment sends at once, built-in destination address translation function (comprising the address camouflage).

These five checkpoints have constituted three links of data flow: input chain [1] → [2], discharging chain [4] → [5] and forwarding chain [1] → [3] → [5].Through the input chain, packet is entered the bag treatment progress of user's space by network interface card; Through discharging chain, packet is sent to external network by this machine process; Through transmitting chain, packet directly is forwarded to the network interface of appointment.

Usually, Netfilter carries out following processing to the network packet of receiving: packet enters system from the left side, carries out after the IP verification, and packet is handled through first checkpoint NF_IP_PRE_ROUTING; Just enter route code then, it determines that this packet is to need forwarding or this machine of issuing; If this packet is what to send out by this machine, then these data pass to upper-layer protocol after handling through checkpoint NF_IP_LOCAL_IN then; If this packet should be forwarded then and be handled by checkpoint NF_IP_FORWARD; After the last checkpoint NF_IP_POST_ROUTING processing of the packet process of transmitting, be transferred on the network again.The local data that produce through checkpoint NF_IP_LOCAL_OUT handle can after, carry out Route Selection and handle, send on the network through NF_IP_POST_ROUTING then.

The Netfilter framework has favorable expansibility, and kernel module can articulate registration to one or more checkpoints, and is called when these checkpoints of packet process, thereby can operate packet at kernel spacing.Do not see that based on the streaming media cluster service content scheduling scheme of Netfilter framework relevant document discloses at present.

Summary of the invention

The objective of the invention is in order to overcome the above-mentioned deficiency of prior art, a kind of streaming media cluster service content dispatching method based on the Netfilter framework is provided.Adopt this method scheduler to carry out unified management,, realized the intelligent scheduling of streaming media service cluster according to the service request that the sheet source distribution and the load state scheduling of server are received to service cluster.

The present invention is based on the streaming media cluster service content dispatching method of Netfilter framework, comprise the steps:

A, user side send service request by network to the virtual address of load balancing cluster;

After b, scheduler are received the requests of packets of data of user side, the activation record that whether has described service request to connect in the dispatch list according to Netfilter framework NF_IP_PRE_ROUTING checkpoint in its (SuSE) Linux OS kernel spacing, if have, the NF_IP_FORWARD checkpoint of described packet by the Netfilter framework directly is forwarded to corresponding destination server; Otherwise, the TCP packet header and the IP packet header of described packet is stored in the temporary table, and described packet is sent into the user's space of scheduler (SuSE) Linux OS by the NF_IP_LOCAL_IN checkpoint of Netfilter framework;

C, at described user's space, scheduler is according to the content of described packet, in program database, find out and store required programs mark server, after taking all factors into consideration the Regional Distribution and loading condition of Servers-all, the destination server of selected this request of response, send described packet the kernel spacing of (SuSE) Linux OS back to by the NF_IP_LOCAL_OUT checkpoint of Netfilter framework then, and in described dispatch list, add this dispatching record;

D, at described kernel spacing, scheduler adopts IP camouflage and IP tunneling to set up user side to be connected to the TCP of destination server, to transmit described packet to destination server;

After e, destination server were received the packet that scheduler transmits, the content of the packet of transmitting according to described scheduler was set up destination server to the TCP response of user side and is connected, and sends required programs to user side.

Wherein, scheduler need be finished following four steps successively in order to help to set up the TCP connection between client and the destination server in steps d:

1., the kernel spacing of scheduler (SuSE) Linux OS is searched in temporary table according to the address and the port information of the user side that is resolved to, find the TCP packet header and the IP packet header of original request, request content is encapsulated again with original T CP packet header and IP packet header, obtain IP bag P1;

2., described kernel spacing is replaced by the address of destination server with the destination address that IP wraps among the P1, and readjusts the header checksum field of IP bag P1, obtains IP bag P2;

3., described kernel spacing wraps P2 with IP and sends to destination server by IP tunnel, be source address promptly with the scheduler address, the destination server address is that destination address encapsulates once more to IP bag P2, obtains IP bag P3, and P3 is sent to the Web by the network interface of appointment;

4., after destination server receives IP bag P3, obtain source address (scheduler address), after client and destination server carry out three-way handshake by scheduler, set up client and connect to the TCP of destination server through one parsing (removing the IP tunnel encapsulation header).

Wherein, when setting up TCP and be connected, user side need carry out following a few step operation at the destination server described in the step e:

(1), after destination server receives the packet that scheduler transmits, carry out twice parsing, obtain the content of the original TCP bag that user side sends;

(2), destination server is obtained the address and the port information of user side from the packet header of original TCP bag;

(3), destination server, is set up destination server and is connected to the TCP of this user side with after corresponding user side carries out three-way handshake according to the address of user side and port information.

The present invention is a kind of method of dispatching based on the streaming media cluster service content of Netfilter framework, utilize scheduler that service cluster is carried out unified management, its scheduler can be taken all factors into consideration the Regional Distribution and the loading condition of sheet source distribution, server, and intelligent scheduling is carried out in request; Be to handle by the (SuSE) Linux OS application layer when user side and backstage video server connect, connect in case after setting up, the subsequent control order can directly be transmitted in the core layer of (SuSE) Linux OS, need not by application layer the system effectiveness height.

With IP camouflage and IP tunneling user's request is transmitted, system's transparency is higher, and when setting up the cluster service system, user side and server program need not to revise.

Description of drawings

Fig. 1 is the exemplary block diagram of load balancing cluster of the present invention system;

Fig. 2 is a Netfilter framework schematic diagram;

Fig. 3 is the delivering path schematic diagram of network packet between user side of the present invention and the server;

Fig. 4 is an overview flow chart of the present invention;

The message format of Fig. 5 TCP bag;

Fig. 6 IP packet voice form.

Embodiment

The present invention is further described below in conjunction with accompanying drawing.

The load balancing cluster system typical structure of realization dispatching method of the present invention as shown in Figure 1, this group system comprises 1 scheduler, some station servers and some user sides, described scheduler, server and user terminal all are connected in the Internet network, scheduler and some station servers adopt star-like topological structure, also can use the network topology of bus-type, other type such as ring-like in the practical application.

As shown in Figure 3, the present invention utilizes the address camouflage to set up a transparent transmission channel between user and server, scheduler is understood the corresponding relation of all user sides and server, user side sends service request to the virtual address of cluster, the scheduler of cluster front end receives that the back is transmitted to background server according to dispatch list with it, and background server directly is handed down to user side with response data packet.All perception is less than the existence of scheduler for whole process user side and server, and handling process is the same when directly carrying out communication with both.

ICP/IP protocol is carried out in network communication of the present invention, and the data cell of application layer is encapsulated in the TCP message of transport layer, and the TCP message is packaged into the IP bag in the process network layer and arrives destination host by Network Transmission.

TCP is a kind of connection-oriented reliable host-host protocol, and it has defined controlling mechanisms such as some verifications, re-transmission, and to guarantee that message can correctly arrive destination host, its message structure as shown in Figure 5.In order between source host end and destination host, to transmit tcp data, must set up a virtual circuit earlier, just TCP connects, and it is as follows that process is set up in the TCP connection:

At first, request end (source host) sends a TCP message that comprises the SYN sign, and SYN promptly synchronous (Synchronize), sync message can indicate the port of source host use and the initial sequence number that TCP connects;

In second step, destination host will return the message of a SYN+ACK after receiving the SYN message of user side, and the request of expression source host is accepted, and the TCP sequence number is added one simultaneously, and ACK promptly confirms (Acknowledgement).

In the 3rd step, source host also returns a confirmation message ACK and gives server end, and same TCP sequence number is added one, connects to this TCP and finishes.Above connection procedure is called as three-way handshake in Transmission Control Protocol.

In transmission course, the affirmation mechanism that also has the similar data bag in the agreement, now between source host and destination host, add a scheduler of transmitting request, do not influence under the prerequisite of original TCP acknowledgement mechanism in assurance, also to not change the configuration of user side and server as much as possible, TCP message of transmitting through scheduler that this just need guarantee that destination host receives and the TCP message when not having the scheduler transfer originally are the same, in other words the TCP that scheduler also will use original user side to send when forwarding packet header of wrapping.Like this, server will be replied according to sequence number and control information in the former TCP packet header, thereby guarantees the reliability that TCP connects between user side and the server.

As shown in Figure 6, the IP bag then comprises source, destination host address, and the check information in IP packet header etc.Because server is according to receiving that the IP address of packet makes response, wanting server directly responds user side, scheduler IP with oneself when transmitting packet carries out the secondary encapsulation to packet, just adopt IP tunneling, address with scheduler encapsulates original IP bag once more as source address, destination server is received after the request package IP bag has been carried out just can obtaining after twice parsing original TCP bag content that user side is sent like this, can set up a server according to the parameter of this TCP bag and data and respond to the TCP of user side and be connected.

The configuration of the present invention's one exemplary embodiments is as follows:

User side (1 is designated as C):

CPU：Intel?PIII?800MHz

Internal memory: 128M

Operating system: Windows 2000 Professional

Player: LSMP Media Player

Dispatch server (1 is designated as D):

CPU：Intel?Xeon?2.4GHz

Internal memory: 512M

Operating system: RedHat8.0 Linux Server

Dispatch server: LSMP Director

Vod server (3, be designated as V1, V2, V3 respectively, be within the same network segment):

CPU：Intel?Xeon?2.4GHz

Internal memory: 1G

Operating system: RedHat8.0 Linux Server

Streaming media server: LSMP Streaming Server

The film source of program S exists only among vod server V1, the V2.

With reference to Fig. 4, concrete scheduling process is described:

1, behind the packet that scheduler D reception user side C sends, carry out agreement and port judges in the NF_IP_PRE_ROUTING checkpoint of Netfilter framework, if host-host protocol is TCP (transmission control protocol) and destination interface is the serve port of VOD (Video on Demand), then enter scheduling flow of the present invention; Otherwise jump out scheduler module, enter the handling process of the original general networks bag of (SuSE) Linux OS.For the packet that enters scheduling flow of the present invention, adopt following steps to carry out the scheduling of service request:

2, scheduler D searches in the dispatch list T of kernel spacing, see if there is the historical record that this asks required connection, if have, the NF_IP_FORWARD checkpoint of this packet by the Netfilter framework directly is forwarded to corresponding server, finish this request scheduling; If do not have this to ask the historical record of required connection in the dispatch list, this packet is sent into the user's space of scheduler (SuSE) Linux OS by the NF_IP_LOCAL_IN checkpoint of Netfilter framework, and the TCP packet header and the IP packet header of this packet is stored in the temporary table;

3, at user's space, scheduler D resolves the packet of receiving, know that the user wishes request program S, search database finds to have among server V1 and the V2 film source of this program, because V1, V2 are within the same network segment,, learn according to the load information that scheduler counts on so when scheduling, only need compare both load sizes, the load of V1 now will be higher than the load of V2, so determine to respond this service request by server V2;

The source that scheduler D will connect, the corresponding relation of destination join among the dispatch list T of kernel spacing;

4, at described kernel spacing, scheduler D utilizes IP camouflage and IP tunneling, sets up TCP and connect between user side C and destination server V2, transmits described packet to destination server.

Scheduler D assists to set up between client and the destination server V2 TCP connection concrete steps and comprises:

After application layer is determined the destination server V2 of this service request of response, kernel spacing is together sent in source address, port, data content and the destination server address of original request, kernel spacing is searched in temporary table according to source address and port, find the TCP packet header and the IP packet header of original request, the data content of request package is encapsulated again with original T CP packet header and IP packet header, obtain IP bag P1 like this;

Then the destination address field (DAF) of P1 is replaced by the address of destination server V2, readjusts the header checksum field of IP bag again, obtain new IP bag P2 like this;

Be source address with scheduler D address again, to be destination address to IP bag P2 encapsulate once more in destination server V2 address obtains IP bag P3, sends to the Web by the network interface of appointment;

Destination server V2 receives that request package removes the encapsulation header of outermost layer IP tunnel earlier, obtains request package P2, and server directly responds client according to the content among the P2.

5, after destination server V2 receives request data package, set up the TCP response connection of destination server V2 according to the content of IP bag to user side C, user side C is connected control channel with destination server V2 by this TCP, in addition, both sides also will consult to create a stream media protocol predetermined data by this TCP connection and issue passage M, and destination server V2 is handed down to the user by described passage M with audio/video flow.

Destination server V2 need carry out following a few step operation when user side C sets up TCP and is connected:

Destination server V2 carries out twice parsing after receiving the IP bag P3 that scheduler D sends, and obtains the content of the original TCP bag that user side C sends;

Destination server V2 obtains the information of user side C from the packet header of original TCP bag;

Destination server V2 with after corresponding user side C carries out three-way handshake, sets up the TCP connection of destination server V2 to user side C according to the user profile that obtains.

Play to finish or user side C sends when stopping request, destination server V2 closes this user thread, stops to issue data flow, and simultaneously, scheduler D will connect from the dispatch list T of NF_IP_PRE_ROUTING and remove.

Claims (4)

1, a kind of streaming media cluster service content dispatching method based on the Netfilter framework is characterized in that comprising the steps:

A, user side send service request by network to the virtual address of load balancing cluster;

After b, scheduler are received the requests of packets of data of user side, the activation record that whether has described service request to connect in the dispatch list according to Netfilter framework NF_IP_PRE_ROUTING checkpoint in its (SuSE) Linux OS kernel spacing, if have, the NF_IP_FORWARD checkpoint of described packet by the Netfilter framework directly is forwarded to corresponding destination server; Otherwise, the TCP packet header and the IP packet header of described packet is stored in the temporary table, and described packet is sent into the user's space of scheduler (SuSE) Linux OS by the NF_IP_LOCAL_IN checkpoint of Netfilter framework;

C, at described user's space, scheduler is according to the content of described packet, in program database, find out and store required programs mark server, after taking all factors into consideration the Regional Distribution and loading condition of Servers-all, the destination server of selected this request of response, send described packet the kernel spacing of (SuSE) Linux OS back to by the NF_IP_LOCAL_OUT checkpoint of Netfilter framework then, and in described dispatch list, add this dispatching record;

D, at described kernel spacing, scheduler adopts IP camouflage and IP tunneling to set up user side to be connected to the TCP of destination server, to transmit described packet to destination server;

After e, destination server were received the packet that scheduler transmits, the content of the packet of transmitting according to described scheduler was set up destination server to the TCP response of user side and is connected, and sends required programs to user side.

2, according to the described streaming media cluster service content dispatching method based on the Netfilter framework of claim 1, it is characterized in that: setting up user side at scheduler described in the steps d to the TCP Connection Step of destination server is:

(1), the kernel spacing of scheduler (SuSE) Linux OS is searched in temporary table according to the address and the port information of the user side that is resolved to, find the TCP packet header and the IP packet header of original request, request content is encapsulated again with original T CP packet header and IP packet header, obtain IP bag P1;

(2), described kernel spacing is replaced by the address of destination server with the destination address that IP wraps among the P1, and readjusts the header checksum field of IP bag P1, obtains IP bag P2;

(3), described kernel spacing wraps P2 with IP and sends to destination server by IP tunnel, be source address promptly with the scheduler address, the destination server address is that destination address encapsulates once more to IP bag P2, obtains IP bag P3, and P3 is sent to the Web by the network interface of appointment;

(4), after destination server receives IP bag P3, obtain the scheduler address, after user side and destination server carry out three-way handshake by scheduler, set up client and connect to the TCP of destination server through one parsing.

3, according to the described streaming media cluster service content dispatching method based on the Netfilter framework of claim 1, it is characterized in that: setting up destination server in step e to the TCP of user side response step of connecting is:

(1), after destination server receives the packet that scheduler transmits, carry out twice parsing, obtain the content of the original TCP bag that user side sends;

(2), destination server is obtained the address and the port information of user side from the packet header of original TCP bag;

(3), destination server, is set up destination server and is connected to the TCP of this user side with after corresponding user side carries out three-way handshake according to the address of user side and port information.

4, according to the described streaming media cluster service content dispatching method of claim 1, it is characterized in that based on the Netfilter framework: record source address, the port of request in the described dispatch list, and the corresponding relation between the destination address, port.

Images