Method for balancing load in multiprocessing unit, and system of multiprocessing unit

Abstract

The method comprises: in term of the size of transmitting data, the data transmitting unit predicts the load of currently-used data process unit, and selects a data process unit with minimum traffic load to transmit data; if there are two data process units both simultaneously having minimum loads, the data transmitting unit selects the one having minimum data size in current cycle to transmit data; if there are two data transmitting units both having minimal data size in current cycle, then using a recurrent selection approach to determine one data transmitting unit to transmit data. The invention can balance the traffic load of data processing unit.

Description

technical field [0001] The invention relates to a multi-processing unit parallel data processing technology, in particular to a multi-processing unit load balancing technology in the parallel data processing process. Background technique [0002] With the wide application of parallel data processing technology, the data distribution technology involved has attracted more and more attention. The following will explain the defects of the existing technology through the description of the softswitch and the data distribution technology in the trunking communication. [0003] In an NGN network, the softswitch device is the core of service processing, processing network protocols such as SIP (Session Initiation Protocol, Session Initiation Protocol), H323, and realizing call and other service functions, such as figure 1 As shown, where UA1 and UA2 are two terminals, they use SIP protocol messages to establish a call through the softswitch device. [0004] The division of softsw...

AI Technical Summary

Problems solved by technology

[0008] Minimum Connection Scheduling / Weighted Minimum Connection Scheduling. This algorithm assumes that when the number of connections of a data processing unit is the smallest, the load is also the smallest. Therefore, the business is distributed to the continuous data processing unit during distribution. However, this high distribution frequency is easy to make The data processing unit receives a large amount of business data in a short time, causing the business load to rise sharply, resulting in overload; especially when it is determined that the processing capacity of the data processing unit is lower than that of other data processing units, although its number of connections is the smallest, the load is not necessarily also the smallest

[0009] Dynamic feedback load balancing algorithm. According to the load feedback of the data processing unit, the algorithm continuously distributes services to the data processing unit with the smallest load during its load query cycle. However, such a high distribution frequency easily makes the data processing unit under load. A large amount of load is received during the query period, and the load rises sharply during the load query period, resulting in overload; especially if the processing capacity of the data processing unit is lower than other data processing units, overload is more likely to occur; and the calculation is more complicated, Not conducive to the application of embedded real-time systems

[0010] To sum up, the existing technology has the following two defects: a certain data processing unit is about to be overloaded, while other units are still relatively idle; or, the load of a certain data processing unit rises sharply within a certain period

Furthermore, the defects of the prior art make it difficult to improve the overall processing capacity of the system

Images