TWM460351U - Cloud computation dynamic work load decision device - Google Patents

Description

Cloud computing dynamic workload decision device

The present invention relates to a workload decision device, and more particularly to a dynamic workload decision device for cloud computing.

Current cloud computing environments are often implemented by a multi-agent system. During the execution of the system, the agent hosts communicate with each other via a communication network and share the workload of the system. Among them, a very important topic of the system is how to dynamically allocate the workload of individual agent hosts.

The workload distribution method currently used in the workload computing device of the cloud computing environment is to use a fixed threshold setting to determine that the workload of the individual agent host is overloaded (overloaded, that is, the workload exceeds the threshold) Or under-loaded (ie, the workload is below this threshold). Then, part of the work of the agent host overloaded by the workload is transferred to a proxy host that is underloaded by the workload.

However, the fixed threshold method has a major drawback in that when the workload of individual agent hosts frequently moves above the threshold, there is a frequent work transfer between the agent hosts, resulting in a ping-pong effect. The phenomenon makes the overall performance of the multi-agent host system low. Therefore, it is necessary to seek a solution.

Therefore, the purpose of the present invention is to provide a dynamic workload decision device for a cloud computing environment, which does not rely on a single fixed threshold to determine the workload of individual agent hosts, and can dynamically allocate such resources dynamically. The workload of the proxy host.

Therefore, the new cloud computing dynamic workload decision device is applicable to a cloud computing environment including a service access point host, a resource allocation host, and a plurality of working hosts, and is electrically connected to the resource allocation host. The host is connected by a communication network, and the service access point host is connected to the Internet to serve as an entrance and exit for the cloud service.

The decision device includes a memory and a processor. The memory is used for storing input explicit value information of a workload of an individual work host collected by the resource allocation host, a fuzzy calculation program module, and output explicit value information output by the fuzzy calculation program module.

The processor is configured to execute the fuzzy computing program module, wherein the fuzzy computing program module comprises a fuzzy unit, an inference unit, and a defuzzification unit. The fuzzification unit is configured to convert the input explicit value information in the memory into at least one fuzzy set by using a attribution function. The inference unit is configured to infer the fuzzy set generated by the fuzzification unit by using the attribution function and a rule base, and generate a fuzzy set of inference results. The defuzzification unit is configured to use the attribution function to convert the fuzzy set of the inference result into an output explicit value information indicating a workload that can be increased or decreased by an individual working host, so that the resource allocation host increases the number of individual working hosts or Reduce the workload.

The effect of the novel model is that, by the design of the fuzzy computing program module, it is possible to infer the output explicit value information indicating the workload that the individual working host can increase or decrease from the input explicit value information of the workload of the individual working host, thereby Increase or decrease the workload for individual work hosts.

1‧‧‧Service Access Point Host

2‧‧‧Resource allocation host

3‧‧‧Working host

7‧‧‧Cloud computing dynamic workload decision device

73‧‧‧Input and output unit (I/O)

74‧‧‧ memory

75‧‧‧ processor

76‧‧‧Fuzzy Computing Module

761‧‧‧Fuzzy unit

762‧‧‧Inference unit

763‧‧‧Defuzzification unit

764‧‧‧ attribution function

765‧‧‧ rule base

Other features and effects of the present invention will be apparent from the following description of the drawings. FIG. 1 is a functional block diagram illustrating the relationship between a preferred embodiment of the present invention and a cloud computing environment. FIG. 2 is a functional block diagram illustrating the functional operation relationship between the components of the fuzzy computing program module of the preferred embodiment; FIG. 3 is a line diagram illustrating the preferred embodiment and the CLB method and the MLB method In comparison, there is less average reaction time; FIG. 4 is a line diagram showing that the preferred embodiment has less average recovery time than the CLB method and the MLB method; and FIG. 5 is a line diagram illustrating the comparison. Compared with the CLB method and the MLB method, the preferred embodiment has more average yield.

Referring to FIG. 1 and FIG. 2, the preferred embodiment of the novel cloud computing dynamic workload decision apparatus 7 is applicable to a cloud computing environment implemented by a multi-agent host system, and may be, for example, a desktop computer or a notebook computer. Tablet or smart phone, etc. The cloud computing environment includes a service access point host 1, a resource allocation host 2, and a plurality of work hosts 3, And the proxy hosts are connected by a communication network. The service access point host 1 connects to the Internet and serves as an gateway for the cloud service. The resource allocation host 2 is used to monitor and control the workload of the working hosts 3. The work hosts 3 are used to perform the tasks required for the cloud service.

In the preferred embodiment, the decision device 7 is electrically coupled to the resource allocation host 2 and includes an input/output unit (I/O) 73, a memory 74, and a processor 75. The memory 74 is configured to store input explicit value information of the workload of the individual work host 3 collected by the resource allocation host 2, a fuzzy calculation program module 76, and the output output by the fuzzy calculation program module 76 is clear Value information. The processor 75 is configured to execute the fuzzy computing program module 76. The input explicit value information includes a processor usage rate and a run-queue length, and the output explicit value information is the number of work tasks that can be increased or decreased.

As shown in FIG. 2, the fuzzy calculation program module 76 includes a blurring unit 761, an inference unit 762, and a defuzzification unit 763. In the fuzzy calculation program, the fuzzy unit 761 first uses the attribution function 764 to convert the processor usage rate and the execution queue length into a fuzzy set corresponding to the overload concept, and a corresponding to the adaptive concept. A fuzzy set, and a fuzzy set corresponding to the underload concept. In the preferred embodiment, the attribution function 764 is a triangle assignment function, but may also be a trapezoidal assignment function, a bell-shaped assignment function, or a Gaussian assignment function.

Then, the inference unit 762 uses the attribution function 764 and a rule base 765 to infer the fuzzy set generated by the fuzzification unit 761 and generate a fuzzy set of inference results. Wherein, the rule base 765 includes : The first rule: If the processor usage is underloaded and the queue length is underloaded, the number of work tasks that can be increased is a large positive integer; the second rule: if the processor usage is underloaded and the execution queue length is overloaded, The number of work tasks that can be increased is about zero; the third rule: if the processor usage is overloaded and the queue length is underloaded, the number of work tasks that can be increased is about zero; the fourth rule: if the processor usage is overloaded And the execution queue length overload, the number of work tasks that can be increased is a small negative integer; the fifth rule: if the processor usage rate is loaded and the queue length is underloaded, the number of work tasks that can be increased is an appropriate amount. Positive integer; sixth rule: if the processor usage is suitable and the execution queue length is suitable, the number of work tasks that can be increased is about zero; the seventh rule: if the processor usage is suitable and the queue length is overloaded , the number of work tasks that can be increased is an appropriate amount of negative integers; the eighth rule: if the processor usage is underloaded and the execution queue length is suitable, the number of work tasks that can be increased is an appropriate amount of positive integers. And ninth rule: If the processor usage is overloaded and execution queue length suitable carrier, you can increase the number of tasks is the right amount of a negative integer. The inference unit 762 uses only the first rule, the second rule, the third rule, and the fourth rule to infer the fuzzy set corresponding to the overload concept and the fuzzy set corresponding to the underload concept. The fuzzy set corresponding to the concept of fit is not inferred, because a work host in a loaded state should not be increased or reduced.

Finally, the defuzzification unit 763 performs defuzzification using a center of gravity and outputs explicit value information indicating the number of work tasks that can be increased or decreased. In the preferred embodiment, the defuzzification unit 763 can also perform the solution by using a mean of maximum, a modified mean of maximum, or a center average. Blurring.

In order to verify the effectiveness of the novel cloud computing dynamic workload decision device 7, the novel creator of the present invention performs experiments according to the present embodiment, and with the current processor-based load balancing (CLB), And memory-based load balancing (MLB) (see Weiwei Lin, James Z. Wang, Chen Liang and Deyu Qi, "A Threshold-based Dynamic Resource Allocation Scheme for Cloud Computing", Procedia Engineering, Vol.23, 2011, pp. 695-703.) A comparison of performance. The experimental setup of the experiment is as follows: the number of basic work tasks of each work host 3 is set to 30K, and each work task is a matrix multiplication; the work originating rate of each work host 3 is λ ₀ =100~2000 (work task/hour), and the general cloud computation density is λ _C = λ ₀ × 0.01 ~ λ ₀ × 1; assume the work task density pattern used for performance analysis (task) The density pattern) is λ _C /λ ₀ , and the delay required for each cloud calculation is limited to 10 seconds; the maximum number of general cloud calculations queued in the first priority category is 15, and is queued in the second priority category. The maximum number of simple cloud computations is 15.

In this experiment, the indicators used for performance evaluation include the reaction time (that is, the time taken from the start of execution to the first response), and the response time (ie, one stroke from the start to the end of execution) The time spent), and the output (ie the number of work tasks completed in a unit of time). As shown in Fig. 3, the present invention has a lower average reaction time than the CLB method and the MLB method under different conditions of the number of tasks. As shown in FIG. 4, the present invention has a lower average recovery time than the CLB method and the MLB method under different conditions of the number of tasks. As shown in FIG. 5, the present invention has a higher average yield than the CLB method and the MLB method under different conditions of the number of tasks.

In summary, with the design of the fuzzy computing program module 76 of the present invention, the number of working tasks that can be increased or decreased by the individual working host 3 can be inferred from the processor usage rate and the execution queue length of the individual working host 3. The resource allocation host 2 is used to increase or decrease the work tasks for the individual work hosts 3. The experimental results show that compared with the current two cloud computing workload decision-making methods, the new model has better performance in response time, response time, and yield evaluation, so it can achieve the purpose of this new model.

However, the above is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention, that is, the simple equivalent changes and modifications made in accordance with the scope of the present patent application and the contents of the patent specification, All remain within the scope of this new patent.

1‧‧‧Service Access Point Host

2‧‧‧Resource allocation host

3‧‧‧Working host

7‧‧‧Cloud computing dynamic workload decision device

73‧‧‧Input and output unit (I/O)

74‧‧‧ memory

75‧‧‧ processor

76‧‧‧Fuzzy Computing Module

Claims (7)

A cloud computing dynamic workload decision device is applicable to a cloud computing environment including a service access point host, a resource allocation host, and a plurality of working hosts, and is electrically connected to the resource allocation host, wherein between the hosts Connected to a communication network, and the service access point host is connected to the Internet for use as a gateway for the cloud service, the decision device includes: a memory for storing the individual work hosts collected by the resource allocation host Input explicit value information of the workload, a fuzzy computing program module, and output explicit value information output by the fuzzy computing program module, wherein the fuzzy computing program module includes a fuzzy unit for utilizing The attribution function converts the input explicit value information in the memory into at least one fuzzy set, and an inference unit uses the attribution function and a rule base to infer the fuzzy set generated by the fuzzy unit and generate an inference result. a fuzzy set, and a defuzzification unit for converting the fuzzy set of the inference result into Show host individual work to increase or decrease the output value of the workload of clear information so that the allocation of resources by the host that increase or decrease the workload of individual staff host; and a processor for executing the fuzzy calculation program modules. The cloud computing dynamic workload decision device of claim 1, wherein the input explicit value information includes a processor usage rate and an execution threshold. a column length, the attribution function respectively converting the processor usage rate and the execution queue length into a fuzzy set corresponding to the overload concept, a fuzzy set corresponding to the fit concept, and a fuzzy set corresponding to the underload concept . The cloud computing dynamic workload decision apparatus according to claim 2, wherein the inference unit only infers the fuzzy set corresponding to the overload concept and the fuzzy set corresponding to the underload concept. The cloud computing dynamic workload decision apparatus of claim 3, wherein the rule base comprises: a first rule: if the processor usage is underloaded and the queue length is underloaded, the number of work tasks that can be increased is one. Large positive integer; second rule: if the processor usage is underloaded and the execution queue length is overloaded, the number of work tasks that can be increased is about zero; the third rule: if the processor usage is overloaded and the queue length is underloaded, The number of work tasks that can be increased is about zero; and the fourth rule: if the processor usage is overloaded and the queue length is overloaded, the number of work tasks that can be increased is a small negative integer. The cloud computing dynamic workload decision apparatus according to claim 1, wherein the defuzzification unit performs defuzzification by using a center of gravity method, a maximum averaging method, a modified maximum averaging method, or a center averaging method. The cloud computing dynamic workload decision apparatus according to claim 1, wherein the output explicit value information of the defuzzification unit is a number of work tasks that can be increased or decreased. The cloud computing dynamic workload decision device according to claim 1, wherein The attribution function is a triangle attribution function, a trapezoidal assignment function, a bell-shaped assignment function, or a Gaussian assignment function.