CN1744593B - A Selection Method of Transmission Link - Google Patents

Abstract

The invention discloses a method for selecting a transmission link. The method separates the load sharing algorithm and the load data management and sets them respectively in the application layer and the transmission layer of the processing unit, and cooperates to complete the distributed application through the application layer and the transmission layer. The processing task distribution process. The method includes: the application layer first detects whether there is a transmission link to the destination processing unit that can process the task, and if it exists, uses the transmission link to complete the task distribution process; otherwise, the application layer selects an appropriate load sharing algorithm, And request an available transmission link from the transport layer through the transport interface; the transport layer uses a load sharing algorithm to select a suitable processing unit from the candidate set of processing units it maintains, and establishes a transmission link to the processing unit; the application layer saves And use the transmission link returned by the transport layer to complete the distribution of processing tasks. The invention realizes the rapid distribution of processing tasks and improves the system scalability of distributed applications.

Description

A Selection Method of Transmission Link

technical field

The invention relates to the communication technology among processing units of a computer distributed application system, in particular to a method for selecting a transmission link when communicating with multiple purpose processing units.

Background technique

With the continuous development of information technology, the scale of computer application systems is expanding rapidly. In order to meet the needs of increasing complexity of computer application systems, componentization and modularization of application programs has become the general trend. Many computer application systems have seen more and more specialized modular division of labor. The complex application program is divided into multiple modules, which are processed by multiple processing units. Each module only focuses on a small aspect of the application system. For example, use a dedicated processing unit to access users, use a dedicated processing unit to manage status data, and so on. Each processing unit communicates with each other through various communication means, combines the processing results, forms an external whole, and provides powerful service functions. The processing unit described here can also be understood as a so-called process.

The above-mentioned pipeline-like operation method greatly improves the scalability of the computer application system, that is, when the performance of a certain processing unit in the computer application system has a bottleneck, as long as the processing capacity of the processing unit is increased, the system requirements can be met. Therefore, , it is convenient to adjust the processing capacity of the application system in real time.

However, the cooperation of each processing unit in the computer application system depends on the support of the communication platform. Therefore, although the above-mentioned operation mode of the pipeline reduces the complexity of the system and improves the scalability of the system, it also increases the communication between the processing units. communication pressure, which may affect the performance of the entire system. In this way, improving the efficiency of cooperation between various processing units will become an important means to improve the overall performance of the computer application system.

Here, how to improve cooperation efficiency includes the following two aspects: first, how to send the cooperation notification to the destination processing unit at the fastest speed; second, how to obtain the fastest processing speed among the processing units that can process the current task. destination processing unit.

For the above-mentioned first problem of improving cooperation efficiency, usually the most common and basic socket (Socket) communication mode can be used to realize the purpose of sending the cooperation notification to the destination processing unit at the fastest speed. Socket can effectively realize fast communication between processes, but the direct application of Socket to realize inter-process communication limits the scalability of the system, the flexibility is poor, and the software reuse ability is not good. For this reason, network programmers usually encapsulate Socket as the basic module of communication in the communication unit, and in the actual process of inter-process communication, directly call the communication unit, so that inter-process communication can be realized flexibly and simply.

Aiming at the above-mentioned second problem of improving cooperation efficiency, that is, to solve the problem of how to select the processing unit with the fastest processing speed among the processing units that can handle the current task, a load sharing mechanism can generally be introduced, that is, before the process communicates, first pass the load The sharing mechanism selects the processing unit with the lightest current load for processing.

The current solutions that effectively combine the above two aspects in distributed applications are generally implemented in a layered form, that is, a transport layer is abstracted to be responsible for managing communication between processing units. The application modules of the entire system will be developed based on the communication support of the transport layer. The transport layer provides communication service interfaces for applications upwards. Through this interface, the application modules use the services provided by the transport layer to complete data transmission between processing units.

Figure 1 shows an abstract model of communication between processing units in a typical computer application system. As can be seen from Figure 1, each processing unit of the model can be abstracted into the following layers of modules: the underlying module, the transport layer and the application layer. Among them, the underlying modules include operating system, middleware, and various common modules; the transport layer is responsible for managing and maintaining the transmission links between processing units, and uses a customized load sharing algorithm to select processing units from the candidate set; the application layer implements business Logic, application transport layer services to complete data communication.

During the operation of the system, the application layer submits a transmission request to the transport layer, the transport layer analyzes the requested task, applies the load sharing algorithm to select a processing unit that can handle this task, and establishes and uses the transmission chain to the processing unit way to complete the transfer process.

It can be seen that in the communication process of the processing units shown in Figure 1, each communication process must first apply the determined load sharing algorithm to select a processing unit from the processing units that can handle this task, and then Data communication is done through the transmission path to the processing unit.

In this way, the time spent in each communication process will include: the time required to retrieve the destination processing unit that can handle this task, and determine the transmission path to the corresponding processing unit; and transmit data on the transmission path The time taken by the package. Here, retrieving and locating the transmission link is only an auxiliary processing unit cooperation, and it is not the necessary time for the distributed system processing unit cooperation.

In addition, the characteristics of a large-scale computer application system are diverse, so applying a predetermined load sharing algorithm at the transport layer does not always work effectively for all communication situations. For example, in the prior art, load sharing is usually randomly selected among the destination processing units that can handle this task through the transport layer. This random selection algorithm generally shows the effect of load averaging of all processing units, that is, the transport layer The task of request processing is evenly distributed among all processing units in the candidate set. Although this load sharing algorithm is simple, easy to implement, and can basically utilize the processing capabilities of each processing unit on an average basis, it has certain limitations in practical applications. For example, when the system dynamically adds a new processing unit or upgrades a certain processing unit, the processing capability of the processing unit, including the central processing unit and memory, may have a large deviation from the original processing unit. In this case, the random selection algorithm obviously cannot fully utilize the processing capability of the newly added or newly upgraded processing unit. This causes the problem that newly added or newly upgraded processing units cannot be quickly integrated with the entire system.

Contents of the invention

In order to solve the above technical problems, the present invention provides a method for selecting a transmission link, which can minimize the time for retrieving processing units and realize rapid processing task distribution.

A further object of the present invention is to more effectively distribute load among processing units and improve the cooperation performance of each processing unit.

The transmission link selection method of the present invention separates the load sharing algorithm processing from the load data management and sets them respectively in the application layer and the transmission layer of the processing unit, and transmits the load sharing algorithm through the transmission interface between the application layer and the transmission layer Information, collaboratively complete the distribution process of processing tasks;

The method includes the following steps:

a. The application layer detects whether there is a transmission link of the processing unit that can process this task, if it exists, use this transmission link to complete the distribution process of this task, and then end; otherwise, execute step b;

b. The application layer selects a suitable load sharing algorithm according to the characteristics of the application program; and requests the transmission layer to provide a suitable transmission link through the transmission interface, and then performs step c;

c. The transport layer judges whether there is a processing unit that can handle this task. If it exists, select an appropriate processing unit from the processing units that can handle this task according to the load sharing algorithm selected by the application layer, and then use the selected The transmission link of the processing unit completes the communication between the processing units; otherwise, it performs exception handling and ends.

The method further includes: after the transmission layer establishes the transmission link and returns the transmission layer link label, the application layer creates the application layer routing label, and establishes an association to the transmission link through a smart pointer.

Moreover, when the state of the transmission link changes, the transport layer modifies the state of the transport layer link mark saved by it; the application layer routing mark automatically learns the change of the state of the transmission link through the smart pointer.

The method for detecting whether there is a transmission link to the processing unit capable of processing this task in step a is specifically: the application layer detects whether the smart pointer of the application layer routing mark is valid.

The selection of a suitable load sharing algorithm according to the characteristics of the application described in step b is specifically: the application program is based on the processor capacity required by the application, or the required memory resources, or the required network bandwidth resources, or any combination of the three Select an appropriate load sharing algorithm.

Step c further includes: after the transport layer selects the destination processing unit, the transport layer first judges whether there is a transmission link to the selected processing unit, and if so, uses the transmission link to complete the communication process between the processing units; Otherwise, the transport layer establishes a transmission link to the selected processing unit, and then uses the established transmission link to complete the communication process between the processing units.

The method further includes: the processing unit receives the load data broadcast by other processing units, and classifies and summarizes the load data of other processing units to generate a processing unit candidate set; in step c, select the target from the processing unit that can handle this task Specifically, the processing unit is: selecting the target processing unit from the processing unit candidate set that can process the current task.

The load data includes the types of tasks that the processing unit can process, the processing capability of the processing unit, and the current load of the processing unit.

The load sharing algorithm described in step c is: the transport layer of the current processing unit searches for the processing unit with the lightest load in all the processing unit candidate sets that can process this task.

The processing unit sets the search step size value before carrying out the load sharing described in step c; and sets a search reference pointer in the processing unit candidate set that can handle this task, pointing to the processing unit at the starting position of the search; The load sharing algorithm described includes the following steps:

b1. In the processing unit candidate set that can handle this task, the processing unit starts from the processing unit pointed to by the search reference pointer, compares the load conditions of the processing units with the number of search step values, and selects the load with the lowest load. light processing unit;

b2. Move the search reference pointer forward by half of the search step value to point to the processing unit at the starting position of the next search.

It can be seen that the application of the transmission link selection method of the present invention can obtain the following beneficial effects:

1. The present invention selects the transmission link saved by the application layer routing label by using the transmission link label when processing task distribution, without repeatedly retrieving the processing unit, thereby reducing the time for transmission link selection and realizing fast The selection of the transmission link, and also avoid the secondary establishment process of the transmission link;

2. The present invention implements the load sharing algorithm function by the application program, and realizes the load data management function at the transport layer, and can flexibly introduce various load sharing algorithms according to the characteristics of the process, thereby more effectively utilizing the processing of each processing unit Capability, improved collaboration performance between processing units, and system scalability for distributed applications.

Description of drawings

Figure 1 shows the abstract model of communication between the processing units of a typical computer application system at present;

Fig. 2 has shown the abstract model that the application system described in a preferred embodiment of the present invention realizes the communication between the processing units;

Fig. 3 is a flow chart of the application layer communication method described in a preferred embodiment of the present invention;

4 is a schematic diagram of the relationship between the processing unit application layer routing mark and the transport layer link mark described in a preferred embodiment of the present invention;

Fig. 5 is a flowchart of a load sharing algorithm according to a preferred embodiment of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

In order to achieve better cooperation among processing units of the distributed application system, a preferred embodiment of the present invention improves the existing load sharing processing method. Fig. 2 shows the abstract model of the communication between the processing units implemented by the computer application system described in this embodiment. As shown in Figure 2, the embodiment separates the load sharing algorithm processing and the load data management and is respectively arranged in the application layer and the transport layer of the processing unit, and between the application layer and the transport layer through a transport interface (TI, Transport Interface) ) transfer the load sharing algorithm information, and cooperate to complete the distribution process of processing tasks.

Wherein, the load sharing algorithm processing module at the application layer can select a suitable load sharing algorithm according to the characteristics of the application itself. For example, if an application has high requirements on processing power and memory resources, the load sharing algorithm designed by the application can select the processing unit with the strongest processing power and the most idle memory resources; If it is relatively high, the load sharing algorithm designed by the application program can select the processing unit with the widest transmission link bandwidth to this processing unit. The load sharing algorithm processing module of the application layer may even consider updating the load sharing algorithm as the load status of the processing units in the candidate set of processing units changes. Load sharing algorithm.

The load data management module located in the transport layer is used to collect and manage the load data of other processing units, for example, the types of tasks that other processing units can handle, the number of tasks that are running, and the maximum processing capacity, etc., and then analyze the load data of other processing units The load data is classified and summarized to form a processing unit candidate set for various processing targets, and generate load indicators for other processing units. Each processing unit candidate set reflects the number and status of processing units that can currently serve a certain processing target in real time.

In this way, through the abstract model described in Figure 2, various limitations in the prior art due to the realization of the load sharing function at the transport layer are changed, and the processing capabilities of each processing unit can be more effectively utilized in the distributed application system. Improve collaboration performance between processing units.

In addition, in order to further reduce the time required for establishing and selecting the transmission link to the destination processing unit and realize fast link selection and communication process, the preferred embodiment of the present invention introduces An application layer routing mark, which is used to record the transmission route from the processing unit to the corresponding destination processing unit, as a basis for whether to reselect the transmission link in the subsequent transmission process.

It will be described below in conjunction with FIG. 2 . If the processing unit 1 establishes a transmission link with the processing unit 2 , the load sharing algorithm processing module of the application layer of the processing unit 1 will save an application layer routing mark to the processing unit 2 . And if the transmission link from processing unit 1 to processing unit 2 is abnormal, the load sharing algorithm processing module of the application layer of processing unit 1 should release the application layer routing label to processing unit 2 . In this way, in the next communication process between processing unit 1 and processing unit 2, the application layer of processing unit 1 can first detect whether the corresponding application layer routing flag is valid, and if the application layer routing flag is valid, then processing unit 1 can directly pass through The established transmission link recorded by the application layer routing mark performs data communication without restarting the load sharing process, thereby saving time for searching and selecting processing units, and realizing fast transmission link selection. Thus, the application program can make full use of the established transmission link for data transmission, without the need to retrieve and select a processing unit during each communication process, thereby realizing fast communication between processes.

Fig. 3 shows the selection process of the transmission link according to a preferred embodiment of the present invention. As shown in Figure 3, the selection process of the transmission link of the present invention includes the following steps:

Step 301: the application program sends a transfer request to the load sharing algorithm processing module of the application layer of the processing unit, and the transfer request includes tasks to be processed by other processing units;

Step 302: The load sharing algorithm processing module detects whether there is an application layer routing mark corresponding to the processing unit that can process the task described in step 301, and if so, executes step 306; otherwise, executes step 303;

Step 303: The load sharing algorithm processing module selects a suitable load sharing algorithm, uses the feature information of this task to request a transmission link from the transport layer through the transport interface; the transport layer judges that there is a corresponding processing unit candidate set, and if so, executes Step 304; otherwise, execute step 307;

Step 304: The transport layer selects a suitable processing unit according to the selected load sharing algorithm, and judges whether there is a transmission link to the selected processing unit, and if so, executes step 305; otherwise, the transport layer establishes a link to the selected processing unit transmission link, then, execute step 305;

Step 305: the load sharing algorithm processing module saves the application layer routing mark, and then executes step 306;

Step 306: The load sharing algorithm processing module uses the transmission link recorded by the application layer routing label to complete the inter-process communication;

Step 307: The application program handles the exception, marking the failure of the current processing task distribution process.

In order to realize the method shown in FIG. 3 , it must be ensured that the application layer routing flag saved by the load sharing algorithm processing module of the processing unit application layer reflects the current state of the transmission link between the processing unit and other processing units in real time. However, due to the hierarchical structure of processing units and the encapsulation principle of communication units in system design, although the transport layer responsible for managing and maintaining transmission links between processing units can obtain the status of transmission links in real time, the transport layer has no direct interface The application layer can be notified that the state of the transport link has changed. In this way, it is very likely that the application layer routing label saved by the load sharing algorithm processing module of the application layer is inconsistent with the actual link state. Obviously, this will lead to the failure of the communication process.

In order to solve this problem, when a transmission link is established in a preferred embodiment of the present invention, the transmission layer of the processing unit also creates a transmission layer link mark to record the transmission link from the processing unit to the corresponding processing unit. Fig. 4 is a schematic diagram of the relationship between the application layer routing label and the transport layer link label of the processing unit described in this preferred embodiment. As shown in FIG. 4 , the transport layer link tag of the processing unit is specifically the connection object of the transport layer, and the application layer routing tag is specifically the routing object of the application layer, and both objects mark the same transport link. Among them, the routing object of the application layer further includes a smart pointer pointing to the connector object of the transport layer. Through the smart pointer, the state of the link mark of the transport layer and the routing mark of the application layer are guaranteed to be consistent, and the transmission link state of the transport layer is hidden from the application layer. The purpose of transparent transmission is that if the smart pointer is not empty, the application layer route is marked as valid; and if the smart pointer is empty, the application layer route is marked as invalid.

When establishing a transmission link at the transport layer, the processing unit establishes a connector object and a routing object at the transport layer and the application layer respectively, and the smart pointer of the routing object at the application layer points to the corresponding transport layer connector object. Both objects mark the transport link to the corresponding processing unit. In this way, when the transmission link is normal, the application program can find the marked transmission link through the smart pointer of its application layer routing object, and complete data communication through the transmission link. When the transmission link is unavailable, the operation of the transport layer to release the corresponding connector object will cause the routing object of the application layer to become invalid. In this way, the application layer routing tag can obtain the status of the corresponding transmission link in real time by detecting its smart pointer.

Thus, the load sharing algorithm processing module can realize the operation of detecting whether the application layer routing mark exists in step 302 by detecting whether the smart pointer of the routing object of the application layer is valid.

In addition, in step 304 of the method of the present invention, the load sharing algorithm processing module can choose the random sharing algorithm in the prior art, and can also choose a variety of other load sharing algorithms. limit.

A preferred embodiment of the present invention provides a load sharing algorithm and a method for managing load data. The load sharing algorithm of this embodiment is suitable for applications requiring high load sharing accuracy. In addition, the load data management in this embodiment utilizes the processing capability of the target processing unit and the load data of the current load situation, etc., and the processing unit with the smallest ratio of actual load to processing capability in the processing unit candidate set, that is, the processing unit with the lightest current load . Wherein, the processing unit candidate set is established and maintained by the payload data management module of the transport layer.

In order to obtain the load status, processing capability and the types of tasks that can be processed by other processing units in real time, each processing unit periodically broadcasts these values to the load data management module of other processing units. The shorter the broadcast period, the higher the transport layer link mark The more accurate the recorded data, the more effective the load sharing algorithm will be. However, on the other hand, the shortening of the broadcast period will lead to an increase in the amount of status information transmitted between processing units, resulting in heavy loads on the network and reduced system collaboration performance. The broadcast period is generally set according to the load information accuracy expected by the application.

It can be seen from the above load sharing algorithm and load data management method that the load data management described in this embodiment does not limit the size of the processing unit candidate set, therefore, new processing units can be continuously added to the system. In addition, because the current load of a newly added processing unit or a newly upgraded processing unit is zero, it will become the main target of processing task allocation for a period of time until the load of this processing unit is gradually equal to the load of other processing units, so This embodiment can ensure that the new processing unit is quickly and seamlessly integrated into the entire distributed system.

However, the time complexity of searching for the lightest-loaded destination processing unit in the entire processing unit candidate set is linear, so the problem of algorithm degradation will appear as the scale of the distributed application system expands. Therefore, in another embodiment of the present invention, its load sharing algorithm adopts a suboptimal search algorithm, that is, the search algorithm no longer searches within the scope of the entire processing unit candidate set, but searches according to the predetermined empirical search step A long value of Step, each time only searches for the number of processing units with the specified step size, and finds a processing unit with the lightest load within the search step as the target processing unit for the distribution of the processing task. This load sharing algorithm is also suitable for the current situation where the load of the task distribution processing unit is high, so that the tasks can be distributed as soon as possible to reduce the occupation of processing capacity.

Fig. 5 is a flow chart of the load sharing algorithm described in this preferred embodiment. As shown in Figure 5, the load sharing algorithm of the present invention sets a search reference pointer in the processing unit candidate set that can handle this task to indicate the starting position of this search, and realizes the task by controlling the movement of the pointer. Each processing unit in the candidate set has an equal chance of being selected; in addition, the algorithm sets an empirical search step value Step, which is set to 5 in the present embodiment; the load sharing algorithm of the present invention includes the following steps:

Step 501: start from the processing unit pointed by the search reference pointer, compare the load conditions of Step processing units, and select a processing unit with the smallest ratio of current load to processing capacity, that is, the processing unit with the lightest load;

Step 502: Move the search reference pointer forward by half of the step value, that is, move Step/2 processing units, as the starting search point for the next search;

Step 503: Establish a transmission link to the processing unit to realize data transmission.

In this embodiment, through multiple load sharing search processes, the search reference pointer can circulate in the candidate set of processing units, so that all processing units in the candidate set of processing units can participate in load sharing, thus ensuring load sharing. The fairness of the algorithm basically realizes the purpose of giving full play to the processing capabilities of all processing units and realizing effective cooperation between each processing unit.

In addition, the load sharing algorithm is to select the processing unit with the lightest load among the preset Step processing units, rather than searching in the entire processing unit candidate set, so that the retrieval time is greatly reduced, thereby improving It improves the efficiency of the application program to search for the destination processing unit.

It should be noted that the processing capacity of the above-mentioned processing units is a virtual processing capacity. This processing capacity may be the nominal index of the system, or it may be a configuration parameter of the system, and it is even possible to implement multiple processing units on one hardware system. In the case of , when the actual processing capacity of the hardware system is allocated to the above-mentioned multiple processing units in a certain way, the processing capacity configured by each processing unit.

It can be seen from the above embodiments that the method of the present invention can flexibly introduce various load sharing algorithms according to the characteristics of the application by moving the processing function of the load sharing algorithm to the application layer of the processing unit and adding the function of load data management at the transport layer. A load sharing algorithm is used to more effectively utilize the processing capacity of the processing unit and improve the cooperation performance between the processing units.

In addition, the method of the present invention can use the transmission link stored in the routing label of the application layer during data transmission by setting up a label for the transmission link, without retrieving the target processing unit twice, thereby reducing the number of selected transmission links. It realizes fast and effective collaboration among processing units.

The preferred embodiments above are cited to further describe the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above descriptions are only preferred embodiments of the present invention, and are not intended to show the present invention. Within the spirit and principles of the present invention, any modifications, equivalent replacements, improvements, etc., shall be included in the protection scope of the present invention.

Claims (10)

1. the method selected of a transmission link, it is characterized in that, this method is with load balancing algorithm process management separates and be arranged at respectively processing unit with load data application layer and transport layer, and between application layer and transport layer, transmitting the load balancing algorithm information by coffret, the distribution procedure of Processing tasks is finished in cooperation;

This method may further comprise the steps:

A, application layer detect whether there is the transmission link that can handle this task handling unit, if exist, then use this transmission link to finish the distribution procedure of this subtask, finish then; Otherwise, execution in step b;

B, application layer are selected suitable load balancing algorithm according to the characteristics of using; And provide suitable transmission link by coffret request transport layer, execution in step c then;

C, transport layer judge whether to exist can handle this task handling unit, if exist, then the load balancing algorithm of selecting according to application layer is selected suitable processing unit from can handle this task handling unit, then, the transmission link that uses selected processing unit is finished the communication between the processing unit; Otherwise, carry out abnormality processing, finish then.

2. the method for claim 1, it is characterized in that, described method further comprises: after transport layer was set up transmission link and returned transport layer link mark, application layer was created the application layer routing label, and was established to the association of this transmission link by intelligent pointer.

3. method as claimed in claim 2 is characterized in that, when the transmission link state changed, transport layer on-the-fly modified its transport layer link flag state of preserving; The application layer routing label is known the variation of this transmission link state automatically by intelligent pointer.

4. as claim 1,2 or 3 described methods, it is characterized in that whether the described detection of step a exists the method for the transmission link that can handle this task handling unit to be specially: whether the intelligent pointer of application layer detection application layer routing label is effective.

5. the method for claim 1, it is characterized in that, the described load balancing algorithm of selecting to be fit to according to the characteristics of using of step b is specially: application program is according to the processor ability of this required by task, or required memory source, or required network bandwidth resources, or three's combination in any is selected suitable load balancing algorithm.

6. the method for claim 1, it is characterized in that, step c further comprises: after described transport layer is selected the purpose processing unit, transport layer at first judges whether to exist the transmission link of selected processing unit, if exist, then use this transmission link to finish communication process between the processing unit; Otherwise transport layer is established to the transmission link of selected processing unit, and the transmission link of use foundation is finished the communication process between the processing unit then.

7. the method for claim 1 is characterized in that, described method further comprises: processing unit receives the load data of other processing units broadcasting, and classifies according to the load data of other processing units and to gather, and generates the processing unit candidate collection;

Step c is described to select the purpose processing unit to be specially from handling this task handling unit: select the purpose processing unit during can handling this task handling unit candidate collection.

8. method as claimed in claim 7 is characterized in that, described load data comprises the disposal ability of the manageable task type of processing unit, processing unit and the present load amount of processing unit.

9. method as claimed in claim 7 is characterized in that, the described load balancing algorithm of step c is: when the transport layer of pretreatment unit can be handled the lightest processing unit of search load in this task handling unit candidate collection at all.

10. method as claimed in claim 7 is characterized in that, processing unit was provided with the step value of search before carrying out the described load balancing of step c; And a search benchmark pointer is set in can handling this task handling unit candidate collection, point to the processing unit that is positioned at the search original position;

The described load balancing algorithm of described step b may further comprise the steps:

B1, this processing unit in can handling this task handling unit candidate collection, from the processing unit of described search benchmark pointer indication, the loading condition of the processing unit of comparison search step value number, and therefrom select the lightest processing unit of load;

B2, described search benchmark pointer is moved forward half of described step-size in search value, as the initial search point of search next time.

Images