CN100435542C - An overload detection method for a communication transaction processing system - Google Patents

Abstract

An overload detection method for a communication transaction processing system is proposed based on the relationship between system resource occupancy rate and response time: high system load will inevitably lead to longer background message queuing time and longer response time. Overload detection method: by measuring the two measurable physical quantities-the service average call arrival rate λ and the corresponding system response time T, and then calculating the resource occupancy rate R according to the relevant formula and mechanism designed by the present invention, Then judge whether the system is overloaded. The method does not depend on various specific environments or implementation conditions of the system, but only automatically performs overload detection according to the business load in the system; and the calculation method is simple, the judgment is accurate, and the work is reliable. This method is combined with the applicant's patent application "Intelligent Network Overload Control Method Based on Service Control Point in Multi-Service Environment", which can well realize the overload detection and overload control of the network system.

Description

An overload detection method for a communication transaction processing system

technical field

The invention relates to an overload detection method for a communication transaction processing system, belonging to the technical field of overload detection in communication systems.

Background technique

It is a very important technical problem to detect the overload of the communication transaction system, because it is the premise of preventing the communication transaction system from being overloaded and implementing flow control. At present, the following commonly used methods are usually adopted for overload detection:

(1) The system using the message queue uses whether the queue overflows as the detection condition. The basis is that when the system load increases, the processing speed slows down, and the message queue waiting to be processed is lengthened. When the queue length increases to a certain value, it is considered overload;

(2) Judging whether the link load is too high, that is, when the link load reaches the warning value of its design capacity, it is considered to be overloaded;

(3) Determine whether the CPU occupancy rate is too high, because the CPU occupancy rate reflects the busyness of the system to a certain extent, and the CPU occupancy rate is used as the overload detection condition, which is effective in a considerable range;

(4) Take the response speed of the system to the message as the overload detection condition.

Although the above-mentioned first three methods can detect and judge whether the system is overloaded within a certain range, they all have certain limitations. Wherein, whether the queue overflows in the method (1) is related to a specific hardware platform, and the length of the overloaded queue is not easy to determine. Method (2) sometimes cannot really reflect the system overload situation, for example, the link capacity is not proportional to the configuration of the processing system. Method (3) is the most intuitive way to judge whether the system is overloaded. However, for a complex distributed system, it is not easy to accurately obtain the CPU occupancy rate, especially the measurement of the CPU occupancy rate can only be performed in actual business In the system, it is usually separated from the space position of the equipment that takes overload control, which will lead to a delay in the transmission of CPU occupancy information, that is, the time for the overload control point to obtain the information of CPU occupancy must lag behind. Therefore, even if the overload control point is adopted The control measures must also lag in time, so that the control effect will be greatly reduced. Moreover, once the occupancy rate information of the CPU is lost during the transmission process, the overload control will completely fail in a serious situation.

Although method (4) is easier to implement than the first three methods, because no matter what causes the overload, the most direct manifestation is that the response speed of the message slows down, that is, it is usually judged whether the system is overloaded by detecting the change of the message response time. is a more effective detection method. However, so far, the method of using response time to detect overload has been based on the pre-measured or set response time value as the reference condition for overload detection; and this method is only applicable to fixed systems and fixed services, that is, for each application to Before the actual application of the system on the existing network, the measurement of the average response time based on the established business must be carried out to obtain the detection benchmark for judging the overload. Otherwise, once the environment changes (including system software and hardware environment changes and/or business environment changes), it will lead to changes in overload detection judgment conditions or benchmarks.

Therefore, how to effectively find the overload detection point of the system and adapt to various changes in the environment has not found a good solution so far, and has become a subject of concern to technical personnel in the industry.

Contents of the invention

In view of this, the purpose of the present invention is to provide a kind of overload detection method for communication transaction processing system, and this method can not depend on various concrete environments or realization conditions of the system, just carry out overload automatically according to the business load size situation in the system detection; and the calculation method is simple, the judgment is accurate, and the work is reliable.

In order to achieve the above object, an overload detection method for a communication transaction processing system is characterized in that: comprising the following steps:

(1) When the system is not overloaded, there is the following relationship between the system response time T and the service average call arrival rate λ: T=a+bλ; in the formula, the parameter a is the transmission time of each message in the system; Parameter b is the proportion factor of the message response time per unit of system resources consumed;

(2) When the system load is low, that is, underload, measure the data of the average call arrival rate λ of multiple groups of services and the corresponding system response time T, and use the least square method to calculate the above two parameters a and b;

(3) In the set time period, measure the service average call arrival rate λ in this time period, and then use the relational expression T=a+bλ in step (1) to calculate the predicted value of the system response time T in this time period T _prediction ;

(4) In the same set time period, measure the actual value of the system response time T in this time period, that is, _{measure the} actual system response time T;

(5) According to the value between the actual response time T _measurement of the system and the predicted value T _prediction of the system response time T, determine whether there is an "overload detection point" in the system:

If T _measurement > T _prediction + ωσ, the system is overloaded; otherwise, the system is not overloaded; where ω is the overload protection factor to prevent misjudgment caused by response time fluctuations, empirical values: 1 to 5; σ is the residual standard deviation.

The method further comprises the steps of:

(6) Dynamically adjust the two parameters a and b on a regular basis, that is, perform step (2) every set time to calculate the values of the two parameters a and b to adapt to changes in the system and business environment.

In described step (2), according to the least squares method, the specific steps of calculating a, b two parameters are:

(21) When the system load is low, measure the data of the average call arrival rate λ _i of N groups of services and the corresponding system response time T _i , i is the serial number of each group of data, N is a positive integer, indicating the measured data group The largest sequence number;

(22) According to the data of average call arrival rate λ _i and system response time T _i of each group of business and the following formula, calculate the following three values respectively:

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;\&lambda;</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ;</span>$

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; TT</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ;</span>$

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;T</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span>$

(23) According to the following two formulas, calculate the two coefficients a and b respectively, and calculate the residual standard deviation σ:

$\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;T</span>}}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;\&lambda;</span>}}}<span\; class="notranslate">\; ;</span>$

$\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; ;</span>$

The formula to calculate the residual standard deviation σ is: $\mathrm{\&sigma;}=\sqrt{\frac{1}{N-2}[S_{\mathrm{TT}}-b_{\mathrm{\&lambda;T}}]}.$

The present invention is an overload detection method (also known as linear regression method) for communication transaction processing system. Data and some calculations can automatically perform overload detection according to the business load in the system; and the calculation method is simple, the judgment is accurate, and the work is reliable. This method is combined with the applicant's patent application "Intelligent Network Overload Control Method Based on Service Control Point in Multi-Service Environment", which can well realize the overload detection and overload control of the intelligent network system. Moreover, the method of the present invention has universality, can be applied to various network systems, and is not limited to the overload detection of intelligent network systems.

Description of drawings

FIG. 1 is a graph showing the relationship between the average response time and the CPU usage (that is, under different loads) in the communication transaction processing system.

Fig. 2 is a flow chart of the operation steps of the overload detection method of the present invention.

Fig. 3 is a test graph of an embodiment of applying the overload detection method of the present invention to modify the adaptive window control algorithm and applying it to the service control point (SCP) system of the intelligent network.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings.

Introduce the working mechanism of the inventive method now: in the transaction processing system, the response time speed of the system to the message is the reflection of the idle and busy degree of the system. It is very good to intuitively understand the change of system load according to the response time. A large number of experimental data analysis results show that under the same load condition, the response time of the system to the message is approximately distributed according to the negative exponential. Under different load conditions, the relationship between the response time of the system to the message and the CPU usage rate is shown in Figure 1.

See Figure 1, which shows the change law of the average response time under different loads: when the system load is light (the CPU occupancy rate is low), the system response time is basically stable and approximately linear; only when the system load is very low When it is heavy, that is, when the CPU usage is greater than 60%, the system response time will increase exponentially.

Research shows that the relationship between system response time T and resource occupancy rate R can be expressed as:

T=c+de ^βR (1)

In the formula, c is a constant, which reflects the processing capability of the system; d is a constant, which reflects the complexity of the message; β is the strength factor, which is usually related to the complexity of the business. In Fig. 1 c=42, d=0.36, β=0.075.

Although the CPU resource occupancy rate and system response time shown in Figure 1 are only based on the test results of a specific platform, they have general significance and are also applicable to other platforms. Although the coefficients of the above formula (1) may be different when running different services on different platforms, the CPU resource usage rate and response time follow the same law: once any system is under high load, it will inevitably lead to background messages. The longer the queuing queue, the response time will increase exponentially.

Usually, in a transaction processing system, the response time T is a measurable physical quantity, and the resource occupancy rate R is a physical quantity that cannot be accurately measured. Especially in a distributed system, the value of R is even more difficult to measure. In the aforementioned formula (1), if the values of the three parameters c, d, and β can be confirmed, and then through the measured value of T, the resource occupancy rate R can be easily calculated according to the formula (1), so as to judge whether the system is overload. Because it is difficult to accurately measure the values of the above three parameters c, d, and β by traditional methods, a convenient and effective method must be found to automatically measure the values of c, d, and β to obtain the functional relationship between T and R. This is also the starting point of the research and development of the present invention.

Since the resource occupancy rate R is a physical quantity that can hardly be measured accurately, it is necessary to find a physical quantity that is easy to measure and can reflect the resource occupancy situation to replace it. After analysis, it is found that this physical quantity exists. The occupancy of system resources mainly depends on the call arrival rate λ of the service, and the call arrival rate λ is a physical quantity that is relatively easy to measure. Therefore, if you can grasp the relationship between λ and R, you can know R indirectly by measuring λ. For the same business, in a statistical sense, each call consumes a certain amount of system resources, and the higher the call arrival rate, the more system resources it consumes; assuming that the system resources are sufficient, λ and R should be a linear relation. Experimental data show that: this linear relationship is established when the system is not overloaded, that is: R = γλ (2)

In the formula, γ is a proportional coefficient, which is related to the specific business. For the convenience of analysis, Taylor expansion of formula (1) is carried out near the coordinate origin 0, and its linear part is taken to obtain the following formula:

T＝c+d+dβR (3)

Substitute formula (2) into formula (3) again to get the following formula:

T=a+bλ (4)

In the formula, a=c+d, b=d ×β×γ, the parameter a represents the time consumed by each group of messages when they are transmitted in the system, including the queuing time in the message queue, the transmission time between different modules, etc. The value is related to the complexity of the message. Parameter b represents the proportion coefficient of each consumed unit of system resources to message response time. Generally, the more complex the message, the more system resources are consumed, which reflects the greater delay of response time.

Based on above-mentioned analysis, overload detection method of the present invention comprises the following operation steps (referring to Fig. 2):

(1) When the system is not overloaded, set the following relationship between the system response time T and the service average call arrival rate λ: T=a+bλ; in the formula, the parameter a represents the time consumed by the transmission of each group of messages in the system Time, including the queuing time in the message queue, and the delivery time between different modules, etc., its numerical value is related to the complexity of the message. The parameter b represents the ratio coefficient of each consumed unit of system resources to the message response time. Generally, the more complex the message is, the more system resources are consumed, and the greater the response time delay will be.

(2) When the system load is low (i.e., underload), collect and measure N groups of system response time T and the corresponding service average call arrival rate λ data, and calculate the two parameters a in the above relational formula according to the least square method , b; the specific steps are:

(21) When the system load is low, measure the data of the average call arrival rate λ _i of N groups of services and the corresponding system response time T _i , i is the serial number of each group of data, N is a positive integer, indicating the measured data group The largest sequence number;

(22) According to the data of average call arrival rate λ _i and system response time T _i of each group of business and the following formula, calculate the following three values respectively:

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;\&lambda;</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ;</span>$

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; TT</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ;</span>$

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;T</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span>$

(23) According to the following two formulas, calculate the two coefficients a and b respectively, and calculate the residual standard deviation σ:

$\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;\; T</span>}}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;\&lambda;</span>}}}<span\; class="notranslate">\; ;</span>$

$\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; ;</span>$

The formula to calculate the residual standard deviation σ is: $\mathrm{\&sigma;}=\sqrt{\frac{1}{N-2}[S_{\mathrm{TT}}-b_{\mathrm{\&lambda;\; T}}]}.$

(3) In the set time period, measure the service average call arrival rate λ in this time period, and then use the formula T=a+bλ to calculate the predicted value _Tprediction of the system response time T in this time period.

(4) In the same set time period, measure the actual value of the system response time T in this time period, that is, _{measure the} actual system response time T.

(5) According to the value between the actual response time T _measurement of the system and the predicted value T _prediction of the system response time T, determine whether there is an "overload detection point" in the system:

If T _measurement > T _prediction + ωσ, the system is overloaded; otherwise, the system is not overloaded; where ω is the overload protection factor to prevent misjudgment caused by response time fluctuations, usually empirical values: 1 to 5; σ is the remaining standard deviation .

(6) Dynamically adjust the two parameters a and b on a regular basis, that is, perform the above step (2) every set time, and calculate the values of the two parameters a and b to adapt to changes in the system and business environment.

In order to verify the effectiveness of the method of the present invention, the method is applied to the Chinese patent application "Control Method of Intelligent Network Overload Based on Service Control Point in Multi-service Environment" (application number: 200510064628.9) for test implementation. Introduce the test situation of this embodiment now:

According to the content in the Chinese patent application "Control Method for Intelligent Network Overload Based on Service Control Point in Multi-Service Environment" (application number: 200510064628.9), the following formula can be obtained:

$\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

In the formula, T is the average response time of the system, λ is the average arrival rate of calls entering the system, and W is the average number of occupied windows. Substituting formula (6) into formula (4), we get:

$\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\sqrt{{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; wxya</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

According to the patent application content of "Control Method of Intelligent Network Overload Based on Service Control Point in Multi-Service Environment", the average occupied window number W of the next measurement point _{can be predicted} according to λ, and then formula (7) can be used Solve the predicted value T _prediction of the system response time T, and then compare the T _prediction value with the actual system response time T _measurement . If the formula T _measurement > T _prediction + ωσ is satisfied, the overload detection point can be determined.

Referring to Fig. 3, this figure is to utilize the self-adaptive window control algorithm after the linear regression method revision of the present invention to be applied to the embodiment test result graph of intelligent network service control point (SCP) system, and its ordinate is the average call arrival rate, The abscissa is the system running time. The hardware platform is a HP ALPHA DS20 minicomputer, within 50 minutes, gradually increase the call arrival rate to 450 calls/second, the protection factor ω=5, judge according to the overload detection method of the present invention, once the overload situation occurs, start immediately overload control. The data of the embodiment is: the optimal window number is 11 when the overload control is started. It can be seen from the figure that under the control of the overload detection method of the present invention, the maximum average call arrival rate received by the SCP system is 196 calls/second, and higher call arrival rates are rejected.

Before the overload detection point is determined, that is, when the average call arrival rate is less than 196 calls/second, each time the linear regression method of the present invention is used to find the overload detection point, the linear relationship between the average response time and the resource occupancy rate in this interval is better , corresponding to the non-overload area in FIG. 1 , it is judged that the overload detection condition has not been met, and no restriction is imposed on the call. But when the average call arrival rate reaches 196 calls/second, the response time and the average call arrival rate are in a non-linear relationship, corresponding to the overload zone of Fig. 1, the overload detection point can be determined according to the formula in the aforementioned step (5), for For subsequent calls, flow control is initiated based on the overload detection condition. In the process of finding the overload detection point, the system adopts the method of the present invention to automatically complete the detection and judgment according to the system load situation. Experimental results show that: the effects of overload detection and overload control are achieved, and the purpose of the invention is realized.

Claims (3)

1. An overload detection method for a communication transaction processing system, characterized in that: comprising the steps of: (1) When the system is not overloaded, there is the following relationship between the system response time T and the service average call arrival rate λ: T=a+bλ; in the formula, the parameter a is the transmission time of each message in the system; Parameter b is the proportion factor of the message response time per unit of system resources consumed; (2) When the system load is low, that is, underload, measure the data of the average call arrival rate λ of multiple groups of services and the corresponding system response time T, and use the least square method to calculate the above two parameters a and b; (3) In the set time period, measure the service average call arrival rate λ in this time period, and then use the relational expression T=a+bλ in step (1) to calculate the predicted value of the system response time T in this time period T _prediction ; (4) In the same set time period, measure the actual value of the system response time T in this time period, that is, _{measure the} actual system response time T; (5) According to the value between the actual response time T _measurement of the system and the predicted value T _prediction of the system response time T, determine whether there is an "overload detection point" in the system: If T _measurement > T _prediction + ωσ, the system is overloaded; otherwise, the system is not overloaded; where ω is the overload protection factor to prevent misjudgment caused by response time fluctuations, empirical values: 1 to 5; σ is the residual standard deviation. 2. The overload detection method for communication transaction processing system according to claim 1, characterized in that: the method further comprises the following steps: (6) Dynamically adjust the two parameters a and b on a regular basis, that is, perform step (2) every set time to calculate the values of the two parameters a and b to adapt to changes in the system and business environment. 3. The overload detection method for communication transaction processing system according to claim 1, characterized in that: in the step (2), the specific steps of calculating the two parameters a and b according to the least square method are: (21) When the system load is low, measure the data of the average call arrival rate λ _i of N groups of services and the corresponding system response time T _i , i is the serial number of each group of data, N is a positive integer, indicating the measured data group The largest sequence number; (22) According to the data of average call arrival rate λ _i and system response time T _i of each group of business and the following formula, calculate the following three values respectively: ${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;\&lambda;</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ;</span>$ ${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; TT</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ;</span>$ ${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;T</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span>$ (23) According to the following two formulas, calculate the two coefficients a and b respectively, and calculate the residual standard deviation σ: $\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;\; T</span>}}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&lambda;\&lambda;</span>}}}<span\; class="notranslate">\; ;</span>$ $\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; ;</span>$ The formula to calculate the residual standard deviation σ is: $\mathrm{\&sigma;}=\sqrt{\frac{1}{N-2}[S_{\mathrm{TT}}-b_{\mathrm{\&lambda;\; T}}]}.$

Images