Method fro dynamic load balance in distributed system based on session origination protocol

Abstract

SIP entity initiates message of SIP request. Receiving SIP message, ILBU sends the message to a PLBU process node based on address or port number of message. Decoding the SIP message, PLBU determines whether the message is a request on inside or outside conversation, or a response in order to carry out different processes. The method improves shortcoming of current fixed type HASH algorithm, and meanwhile keeps characteristic of no state of distributing procedure self. In the invention, no meter increasing or decreasing nodes in the system or adjusting distributing strategy of system will not influence on processing message of conversation so as to raise reliability and scalability of system greatly.

Description

technical field [0001] The invention relates to load balancing in a SIP (SESSION INITIATION PROTOCOL, session initiation protocol) distributed system. Background technique [0002] SIP is a kind of multimedia communication control protocol, which has been widely used in networks such as softswitch and IMS (IP Multimedia Subsystem, IP Multimedia Subsystem). SIP adopts text encoding method and uses tracking analysis, but at the same time, it also causes the message volume to be too large, which brings a heavy burden to the SIP server at the core. [0003] Large-scale systems generally use distributed system processing. For servers without call status, the load balancing is relatively simple, but for SIP servers with call status, such as SoftSwitch and CSCF (CALL SESSION CONTROL FUNCTION, call session control function), the load Equilibrium is a complex issue. Before analyzing SIP distributed processing, we can divide SIP distributed processing nodes into three logical entiti...

AI Technical Summary

Problems solved by technology

[0005] One is user-based distributed processing, that is, user IDs are used as keywords for load balancing. Its disadvantage is that the system load balancing components must store user status. When the system capacity is large, the entrance load balancing components often Multiple nodes are required, and the overhead of synchronizing state between multiple nodes is too large

[0006] The second is to use different IP addresses or transport layer ports to identify different processing node numbers. For example, port 5060 is used as the well-known port of the SIP node, and each node is assigned a unique port. When the SIP message passes through the node, according to The internal complex balancing algorithm allocates a node for it, and then the node performs the Record-Route operation to record the port number of the node, and the other network element will send the request according to the new port after receiving 200 OK. The disadvantage of this method is that when using When the connection is transmitted, the upstream SIP network element must re-establish the connection with the local end, which increases the underlying overhead

At the same time, the disadvantages of this solution are also obvious. Since the call processing is stateful, the same call must be processed in one node. When a node is added to the system or a node is reduced due to a failure, the HASH modulus changes, and the established call Subsequent SIP messages cannot be guaranteed to be distributed to the same node for processing, nor can dynamic load balancing be achieved

Such as a call Call-ID=C1, HASH(C1)=18, the total number of nodes in the system is 5, then the processing node of the call is 18Mod5+1=4, and the system adds a node in the call holding stage, the total node number becomes 6, then the follow-up message is wrongly sent to node 1 (18Mod6+1) (because it is assumed that the node number starts from 1, and 1 needs to be added after taking the modulus, 18mod6=0), so that the call cannot be continued or released normally

In the actual system, the distributor may adopt a certain load balancing strategy, but the above defects still exist under the condition that each call state is not stored

Images