KR101400500B1 - Method and appratus for stochastic scheduling in differentiated service system - Google Patents

Abstract

The present invention relates to a scheduling apparatus and method in a self-similar traffic differentiation service system.

A scheduling method for providing a differentiated service in a packet communication system provided by the present invention includes the steps of: checking whether current traffic is the self-similar traffic; if the current traffic is self-similar traffic, Searching for a target class that satisfies a quality of service (QoS) of the current traffic when the class moving condition is satisfied; Changing the source class to the target class, and changing the source class to the highest class among a plurality of predetermined classes when the target class is not searched.

DiffServ, statistical analysis, Quality of Service (QoS), scheduling

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a statistical scheduling apparatus and a scheduling method in a differential service system,

The present invention relates to a scheduling apparatus and method in a differentiated service system. More particularly, to a scheduling apparatus and method in a self-similar traffic differentiation service system.

Self-similar traffic refers to statistical characteristics of arbitrary intervals in the signal and similar statistical characteristics in the sub-interval. It generally refers to the fact that the whole shape and the shape of the part are repeated in a similar fashion, as can be seen in the fractal image. In the design of the network in the existing communication system, the traffic is assumed to have the characteristics of a mathematical model such as a Poisson distribution or an exponential distribution. However, it has been found that the measured traffic actually has different characteristics from the mathematical model. Studies have shown that many of the traffic generated in communication systems are known to have self-similarity.

At the beginning of the development of the Internet, the purpose was achieved only by transmitting information between the end points such as e-mail and web service. In this way, the Internet has merely had to simply 'transfer' information on the Internet. Thus, the Internet network has a simple characteristic of transmitting information, delaying delivery or loss depending on whether or not the router has communication resources. However, recently, VoIP (Voice over IP) and real-time broadcasting combined with the Internet have led to the situation that the Internet network must not only transmit information but deliver it at a precise time. In addition, the required level of quality varies depending on the service.

In order to support application services having various requirements as described above, there is a need for many techniques to meet not only an increase in bandwidth but also reliability and real-time transmission. These technologies are collectively referred to as QoS (Quality of Service) management technology. Hereinafter, a differentiated service, which is a representative QoS management model, will be described.

The Internet Engineering Task Force (IETF), a standardization organization for Internet-related technologies, presents an end-to-end QoS management model for user traffic on the Internet. Typical examples are Integrated Service (IntServ) and DiffServ (DiffServ). The integrated service is a method of processing real-time traffic and traffic having an upper class first, and a differentiation service is a method of processing different kinds of traffic.

Integrated Services (InterServ) are known to require a large amount of memory space for maintaining and managing the state of numerous individual packet flows and increasing the processing load. In addition, all routers in the network must have the functions of ReSource Reservation Protocol (RSVP), admission control, packet classification, and packet scheduling. In other words, it has a disadvantage in scalability because it requires maintenance of a wide range of status information over the whole network of the Internet. Because of this problem, it is considered that integrated services are difficult to use universally.

The differentiation service differentiates services by a set of packet flows of a real-time application service, and can easily apply various services with scalability in a large-scale network. In addition, the differentiated service is different from the concept of providing different QoS (Quality of Service) for each IP packet flow, that is, it provides different QoS by a certain aggregation as a unit, Such as DSCP recording that requires the router to be located at the edge of the network (the router at the border between the networks is called the edge router) Core) routers have a very simple packet forwarding capability and are scalable so that they can be applied to a large-scale Internet network connected to a plurality of Internet Information Providers (ISP) networks.

The differentiated service guarantees QoS for each classified aggregation, unlike the integrated service model that guarantees QoS for individual flows. Therefore, there is little processing load and state information to be maintained at the node, and the complex traffic control function is designed to be performed only at the edge router. Therefore, it is easy to implement compared to the integrated service using RSVP (resource reservation protocol) by performing simple packet forwarding function in core router. With this advantage, differentiated services are attracting attention, and scheduling methods for providing QoS in differentiated services are being actively researched.

FIG. 1 is a block diagram of a general IP packet used for the differentiating service.

The differentiation service uses a ToS (Type of Service) field consisting of 8 bits in a conventional IP packet. That is, the first 6 bits of the conventional ToS field are used as a DSCP (DiffServ Code Point) field, and the remaining 2 bits are not used at present. The DSCP field indicates a class of a service to which the packet belongs. That is, the DSCP field is used to indicate the QoS that should be provided to the packet when it passes through the router. Table 1 below is a table defining classes according to the priority of the differentiated services.

Class 1 is the service with the highest priority, and class 2 is the service with the second highest priority. It is needless to say that the criterion of which class each service belongs to can vary depending on the setting of the service provider. The differentiation service marks the class number indicated in Table 1 as a positive integer in the DSCP field of the IP packet to indicate the priority of the corresponding packet.

Hereinafter, a scheduling method for QoS management will be described.

In a packet communication system, in order for a plurality of source generated packets to be transmitted from one node to a destination address of each packet, a rule for processing packets in the node is required. These rules are called scheduling. That is, the scheduling technique refers to a queuing process that determines which packet should be processed first among a series of packets waiting in one or a plurality of queues in order to send from a device such as a router in a packet communication system to a next link . Hereinafter, a general scheduling method will be briefly described.

First, there is a first input first out (FIFO) scheme. The FIFO scheme processes packets according to the order in which they arrive at the queue. That is, the packets arriving at the node among the packets generated in the plurality of streams are processed first. The FIFO method is widely used because it is simple to implement. However, if there is no more space in the queue, new packets are discarded and packet loss occurs. Therefore, it is not suitable for use in services sensitive to packet loss.

The second is fair queuing (FQ). The FQ scheme is a scheme for fairly processing a plurality of queues into which a plurality of streams are input. Although there is an advantage in fairness, when a packet having a large size is input in one stream, there is a problem in that processing for another stream is delayed.

The third is the Priority Queuing (PS) scheme. The PQ method is a modified version of the FIFO queuing method. It separates traffic of a specific type and sends it to the front part of the output so that it can be processed first. In this method, the more service differentiation steps are performed, the more processing burden is imposed and the packet forwarding performance is deteriorated. When the amount of the highest priority traffic is large, the low rank traffic increases the loss rate due to the buffer depletion, and the delay-sensitive service may not operate properly. In an increasingly fast network environment, this queuing method is difficult to support scalability.

 The fourth is weighted fair queuing (WFQ). The WFQ scheme is based on the fairness of controlling traffic with queues on a per-stream basis so that a small amount of traffic is not damaged by a large amount of traffic, and a weight based on a specific criterion, It is a combination of weighting aspects that discriminate. Currently, it is the most used method for scheduling of differentiated services.

However, in the above scheduling schemes used in the differentiating service, the class of traffic once set is fixed. Therefore, if some of the traffic corresponding to a certain class is traffic requiring QoS higher than QoS guaranteed by any class, the traffic will be processed without satisfying the QoS, and there is a high probability of data loss.

In order to solve the above problems, one aspect of the present invention provides a method and apparatus for guaranteeing QoS in a differentiated service.

Another aspect of the present invention provides a method and apparatus for efficiently using resources in a differentiating service.

Another aspect of the present invention provides a method and apparatus for preventing data loss in a differentiating service.

A scheduling method for providing a differentiated service in a packet communication system provided by the present invention comprises the steps of: checking whether a current traffic is a self-similar traffic; if the current traffic is the self-similar traffic, Searching for a target class that satisfies a quality of service (QoS) of the current traffic if the class moving condition is satisfied; and if the target class is found, Changing the source class to the target class, and changing the source class to the highest class among a plurality of predetermined classes when the target class is not searched.

The present invention also provides a scheduling apparatus for providing a differentiated service in a packet communication system, the method comprising: checking whether a current traffic is a self-similar traffic; if the current traffic is the self-similar traffic, Searching for a target class that satisfies a quality of service (QoS) of the current traffic if the class moving condition is satisfied, and if the target class is found, And changing a source class to a highest class among a plurality of predetermined classes when the target class is not searched.

Effects of the present invention will be briefly described as follows.

The QoS of the traffic can be guaranteed by preventing data loss of the traffic inputted to the scheduler in real time. And by class moving for specific traffic belonging to a class, QoS can be guaranteed for traffic while efficiently using communication resources. Statistical analysis can further refine traffic processing by classifying classes more precisely, thus further diversifying the quality of communication services.

The operation principle of the preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. The following terms are defined in consideration of the functions of the present invention, and may be changed according to the intentions or customs of the user, the operator, and the like. Therefore, the definition should be based on the contents throughout this specification.

The present invention determines the boundary value of the queue length for providing QoS for each traffic class by statistical analysis. After that, if the traffic inputted to the scheduler corresponds to a certain condition, the traffic class is changed from the originally set class to another class to provide the appropriate QoS.

2 is a configuration diagram of a scheduling system according to an embodiment of the present invention. The scheduling system of the present invention includes a priority classifier 201, queues 211, and a scheduler 221. It is assumed that a plurality of classes are set for the differentiated service traffic in the present invention. In the present invention, each class may not consist of only the same type of service (for example, VoIP) but may be configured of packets requiring a similar QoS even for services of the same type.

The priority classifier 201 classifies input traffic into classes and outputs the classified traffic to a queue 211. That is, packets input from a plurality of lines are sorted by packets having similar QoS, and are output to queues 221 for processing each class. The queues 221 output packets to the scheduler 221 for each class. The internal structure of the scheduler 221 according to the embodiment of the present invention will be described separately in FIG. 5 which will be described later.

Prior to the description of the flowchart of the method of the present invention, the method of determining the range (upper limit value and lower limit value) by class according to the statistical analysis proposed by the present invention will be described first.

The boundary value of the average queue length by statistical analysis is determined in the following order.

는 class j 트래픽이 스케쥴러에 입력되는 시간당 패킷 수(

,이하 같다.)(One)

Is the number of packets per second the class j traffic is input to the scheduler (

, The same shall apply hereinafter.)

It is assumed that the class j is the class to which the current traffic belongs.

은 Class j 트래픽에 할당되는 서버의 용량이고 하기 <수학식 1>로 구할 수 있다.(2)

Is the capacity of the server to be allocated to the Class j traffic and can be obtained by Equation (1) below.

는 패킷들이 스케쥴러에 실시간으로 도달할 때, 임의의 기준 시간에 대하여 n 번째 시간 동안의 서버의 전체용량에 해당한다.At this time

Corresponds to the total capacity of the server for the nth time for any reference time when the packets arrive at the scheduler in real time.

은 n번째 시간의 class j트래픽의 큐의 길이이고 하기 <수학식 2>로 계산된다.(3)

Is the length of the queue of the class j traffic of the n-th time and is calculated by Equation (2) below.

(class j의 트래픽에 대해 n번째 시간동안 새로 스케쥴러에 도착하는 패킷의 수)와,

(서버의 용량)을 감안하여 린들리(Lindley)방정식을 확장하여 계산되었다. 상기 린들리 방정식에 대한 설명은 <http://en.wikipedia.org/wiki/Lindley_equation>에서 참고할 수 있다.In Equation (2), the length of the queue is

(the number of packets arriving at the new scheduler for the nth time for traffic of class j)

(The capacity of the server), and extended the Lindley equation. A description of the Lindley equation is available at http://en.wikipedia.org/wiki/Lindley_equation.

는 n번째 시간 동안 class j 트래픽을 처리할 수 있는 서버의 잔여 용량이고 하기 <수학식 3>으로 계산된다.(4)

Is the remaining capacity of the server that can process class j traffic for the n-th time period and is calculated by Equation (3) below.

는

의 표준 편차에 따른 서버(server)에서 class j 트래픽의 평균 큐 길이로 정의되며 하기 <수학식 4>로 계산된다.(5)

The

Is defined as an average queue length of class j traffic in a server according to the standard deviation of the class j traffic.

과

이 서로 상관성이 적은 특성이 있다. 이 특성을 이용하여 상기 <수학식 4>는 얻어질 수 있다. In Equation (3)

and

There is a characteristic that there is little correlation with each other. Using this characteristic, Equation (4) can be obtained.

은 class j 패킷들의 표준편차를 의미한다. remind,

Is the standard deviation of class j packets.

는

의 상한값(Upper bound)을 의미하며 하기 <수학식 5> 또는 <수학식 6>으로 구해진다.(6)

The

And is obtained by the following equation (5) or (6). &Quot; (6) &quot;

을 이용하여 구할 수 있다. Equation (5)

. &Lt; / RTI >

)는 class j의 잔여의 서버 용량 (

)와 서로 낮은 상관성을 가진 것으로 알려져 있다. 이 성질을 이용하면

로 가정할 수 있다. 상기 가정을 이용하면 상기 <수학식 5>는 하기 <수학식 6>으로 다시 쓸 수 있다.Also, the number of packets in class j (

) Is the remaining server capacity of class j (

) Are known to have low correlation with each other. Using this property

. Using the above assumption, Equation (5) can be rewritten as Equation (6) below.

는

의 하한값(Lower bound)를 의미하며 하기 <수학식 7>로 구해진다.(7)

The

And is obtained by the following equation (7). &Quot; (7) &quot;

또는

의 성질을 이용하여 구할 수 있다. 여기서

은

의 분산(variance)을 의미하며,

는

의 분산을 의미한다. Equation (7)

or

Can be obtained by using the properties of. here

silver

, And a variance of < RTI ID = 0.0 >

The

.

의 상한값과 하한값을 구하는 과정을 보였다.Through the processes of (1) to (7)

And the upper limit and the lower limit of the value.

의 범위를 다시 표현하면 하기 <수학식 8> 또는 <수학식 9>와 같이 표현된다.The above-described process is summarized

The range of expression (8) can be expressed as Equation (8) or Equation (9) below.

(트래픽의 시간당 도달률),

(전체 서버 용량),

(서버의 잔여 용량)인 것을 알 수 있다. 즉, 상기의 인자들의 통계적 분석을 통하여 해당 클래스 트래픽의 평균 큐 길이를 추정하는 것이다. 도 3은 본 발명의 실시예에 따른 차등화 서비스의 클래스별 트래픽의 평균 큐 길이를 추정하는 예시도이다. 도 3에서

에 따라 5개의 클래스별로 평균 큐 길이(

)의 상한값과 하한값이 결정된 것을 볼 수 있다. 상위 클래스일수록

의 범위가 좁아지는 것을 볼 수가 있다.So far, the average queue length of traffic according to class has been defined by statistical analysis. The factor by which this is determined is

(Traffic per hour),

(Total server capacity),

(The remaining capacity of the server). That is, the average queue length of the class traffic is estimated through a statistical analysis of the above factors. 3 is an exemplary diagram for estimating the average queue length of traffic for each class of the differentiating service according to the embodiment of the present invention. 3,

Average queue length per class of 5

The upper limit value and the lower limit value are determined. Higher classes

Can be seen to be narrowed.

4 is a flowchart illustrating a scheduling method for providing a differentiating service according to an embodiment of the present invention. The flowchart of FIG. 4 assumes a state in which the upper and lower limits of the average queue length of each class are determined by Equation (9).

In step 401, the scheduler checks whether the traffic is a self-similar traffic. If the traffic is not self-similar traffic, the conventional scheduling method is used in step 411, and if the traffic is self-similar traffic, the performance of the system can be increased by using a scheduling method of step 403 or less.

In step 403, the scheduler analyzes the incoming traffic in real time to check whether it corresponds to class moving condition 1 or 2, which will be described later. The class movement condition refers to a condition that the length of the average queue of the current traffic approaches a risk level of the QoS level and thus the QoS is not provided.

In step 405, the scheduler searches for a target class. The target class refers to a class that is changed in a class to which current traffic belongs (hereinafter referred to as an initial class). The conditions for searching for the target class will be described later. If the target class is found in step 405, the process proceeds to step 407.

In step 407, the class of the current traffic is changed from the source class to the target class. The scheduler then processes the current traffic class as a target class and provides QoS to the current traffic. If the target class is not found in step 405, the flow advances to step 409.

In step 409, the current traffic class is changed from the original class to the highest class. The highest class is a class set to receive the best QoS in the current network. The absence of a target class means that there is no class that can provide adequate QoS for the current traffic. Therefore, the scheduler changes the primitive class to the top-level class. This is because it is the best way to provide the best QoS for current traffic in the current network.

The process described in FIG. 4 will be briefly summarized as follows. The average queue length per class is determined by statistical analysis in advance. Thereafter, the traffic input to the current scheduler is analyzed in real time, and if it corresponds to the class moving condition, the target class that satisfies the QoS of the current traffic is searched. When the target class is found, the current traffic is changed to the target class and processed. If not, the class of the current traffic is changed to the top class and processed.

Although not described in FIG. 4, when the current traffic corresponds to a class moving condition (that is, a risk level at which QoS can not be provided) in step 403, in addition to moving the class of the current traffic, A method of increasing the processing capacity of the server allocated to the server may be used. A further detailed description thereof will be omitted.

Hereinafter, the class moving condition described in step 403 will be described in detail.

,

,

에 대하여 설명한다.First, the arguments used in the class moving condition

,

,

Will be described.

는 class j 트래픽에 허용 가능한 큐 길이의 최대값이다. 이것은 미리 결정되는 값이다.

Is the maximum allowable queue length for class j traffic. This is a predetermined value.

는

의 Threshold 1로 하기 <수학식 10>으로 표현된다.

The

Is expressed as < EMI ID = 10.0 >

)가 앞서 설명한 상한값(

)에 근접한 정도를 나타낸다. 이 때

값은 스케쥴러에 미리 설정되어 있어야 한다. 상기

는

가

에 근접한 정도를 나타내는 척도이다. 만약

값이 0.95로 설정되었다면, 스케쥴러는

가

의 95%에 해당하는 값을 Threshold 1으로 사용하여 제어한다는 의미이다.

는 시스템의 운영 방식, 트래픽의 특성 및 클래스의 우선 순위(priority)에 따라 변경될 수 있음은 물론이다.Equation (10) represents the average queue length of currently input traffic

) Exceeds the upper limit value (

). &Lt; / RTI > At this time

The value must be preset in the scheduler. remind

The

end

And the like. if

If the value is set to 0.95, the scheduler

end

Is used as the threshold value 1 to control the value corresponding to 95% of the threshold value.

May be changed according to the operating system of the system, the characteristics of the traffic, and the priority of the class.

가

에 근접한 정도를 알아야 하는 이유는 다음과 같다. The scheduler

end

The reason why you need to know the degree of proximity is as follows.

)가 상한(

)에 근접할수록 현재 트래픽의 표준편차(

)가 약간만 변화하여도 큐 길이의 증가는 급격하게 나타나는 성질이 있다. 만약 큐 길이가 급격하게 증가하여 큐 길이 최대값(

)을 초과한다면 이후 스케쥴러에 입력되는 패킷들은 버려져서 패킷 손실이 발생하는 문제가 생길 수 있다. 따라서 현재의 큐 길이가

에 근접하고 있다면 후술되는 클래스 이동 조건에 따라 클래스 이동 제어를 미리 수행하여 안정적으로 QoS를 제공할 수 있다. The current average queue length (

) Is upper limit (

), The standard deviation of the current traffic (

) Is slightly changed, the increase of the queue length becomes sharp. If the queue length increases drastically, the maximum queue length value (

), Packets to be input to the scheduler may be discarded, resulting in packet loss. Therefore, if the current queue length is

It is possible to stably provide QoS by performing class moving control in advance according to the class moving condition described later.

는

의 Threshold 2로 하기 <수학식 11>로 표현된다.

The

Is expressed as < EMI ID = 11.0 >

)가 트래픽에 허용 가능한 큐 길이의 최대값(

)에 근접한 정도를 나타낸다.Equation (11) represents the average queue length of currently input traffic

) Is the maximum allowed queue length for traffic (

). &Lt; / RTI >

값은 스케쥴러에 미리 설정되어 있어야 한다.

는

(현재 트래픽의 평균 큐 길이)가

(class j 트래픽에 허용 가능한 큐 길이의 최대값)에 근접한 정도를 나타내는 척도이다. 만약

값이 0.9라면, 스케쥴러는

가

값의 90%에 해당하는 값을 Threshold 2로 사용하여 제어한다는 것을 의미한다.

값은 시스템의 운영 방식, 트래픽의 특성 및 클래스의 우선 순위(priority)에 따라 변경될 수 있음은 물론이다. At this time

The value must be preset in the scheduler.

The

(The average queue length of the current traffic)

(the maximum value of the allowable queue length for class j traffic). if

If the value is 0.9, then the scheduler

end

Means that the value corresponding to 90% of the value is used as Threshold 2 to control.

It goes without saying that the value may be changed according to the operating method of the system, the characteristics of the traffic, and the priority of the class.

가

에 근접한 정도를 알아야 하는 이유는 다음과 같다.The scheduler

end

The reason why you need to know the degree of proximity is as follows.

)가 허용된 큐 길이 최대값(

)을 초과하는 순간 큐 길이 위배(queue length violation)가 발생하기 때문이다. 큐잉(Queueing) 이론에 따르면 상기 큐 길이 위배는 QoS 위배로 해석되는 것으로 알 려져 있다. 따라서

가

에 근접했을 때 미리 클래스 이동 제어를 수행하여 현재 트래픽에 안정적으로 QoS를 제공할 수 있다.Average queue length for current traffic (

) Is the maximum queue length allowed (

), The queue length violation occurs. According to the queuing theory, it is known that the queue length violation is interpreted as a QoS violation. therefore

end

It is possible to stably provide QoS to the current traffic by performing class moving control in advance.

,

,

에 의하여 클래스 이동 조건을 상세히 설명한다. 본 발명의 클래스 이동 조건은 두 가지 경우이다.The aforementioned factors, i. E.

,

,

The class moving condition will be described in detail. The class moving condition of the present invention is two cases.

(1) Class movement condition 1

는 미리 설정된 값이다.

가

인 이유는 다음과 같다. 스케쥴러가 현재 트래픽의 평균 큐 길이(

)가 상한값에 가까이 있는 경우 (즉,

이고, 0<

<1이므로

는 상한값(

)에 가까이 있게 된다.)

의 작은 변화에도 급격하게 큐 길이가 증가될 수 있기 때문에 클래스 이동 제어를 미리(

이고

이므로

에 도달하는 시간이 Threshold 2(

)에 도달하는 시간보다 '미리'가 된다.) 수행하여 현재 트래픽에 안정적으로 QoS를 제공하는 것이다.

Is a preset value.

end

The reason for this is as follows. The scheduler determines the average queue length of the current traffic (

) Is close to the upper limit value (i.e.,

, 0 <

<1

Is the upper limit value

).

The queue length can be rapidly increased even in a small change of

ego

Because of

Is less than Threshold 2 (

) Of the current traffic, and provides QoS stably to the current traffic.

(2) Class movement condition 2

If the current traffic satisfies the class moving condition 1 or 2, the class moving control is performed. However, in this case, the traffic receiving the class mobility control is not the entire traffic belonging to the class j. The controlled traffic satisfies a certain criterion among class j traffic. That is, the class movement control may be performed on traffic corresponding to a specific condition set by the system through the headers (tagged header, label header) indicating the transmission and reception addresses of the traffic and the priority of the class . It goes without saying that the specific conditions may vary depending on the system.

Hereinafter, the conditions for searching for the target class in step 405 will be described in detail.

If the current traffic belongs to class j and corresponds to the class moving condition, the scheduler searches for a target class that can satisfy the QoS of the current traffic. Since the current traffic is class j, we will search target classes sequentially for class j-1 to class 1. That is, it checks whether a class having a higher priority than the class to which the current traffic belongs corresponds to the target class. If not, then the next higher priority class is checked for the target class. This process continues until the top-level class is found until the target class is found.

In the present invention, the conditions of the target class are set as follows.

and <조건 2>

and <조건 3>

<Condition 1>

and <condition 2>

and <condition 3>

If the class under examination (hereinafter referred to as &quot; inspection class &quot;) satisfies all of the above <condition 1> to <condition 3>, the class is determined as the target class. Hereinafter, the meanings of <Condition 1> to <Condition 3> will be described in detail.

에서

는 검사 클래스의 큐 길이의 하한(Lower bound)을 의미한다. <Condition 1>

in

Means the lower bound of the queue length of the check class.

는 앞서 설명한 것처럼 class j에 설정될 수 있는 큐 길이 최대값이다. 상기 <조건 1>의 의미는 다음과 같다. 현재 트래픽의 큐의 길이는 검사 클래스의 경계값 내에 속해야 한다. 그런데 클래스에 따라서는 경계값의 간격이 매우 좁은 경우가 있다. 이 경우에는 검사 클래스의 큐 길이의 하한이 현재 트래픽의 평균 큐의 길이보다 큰 경우가 생길 수 있다. 현재 트래픽의 큐의 길이가 검사 클래스의 큐 길이의 범위에 속하지 못할 수 있다. 이렇게 되면 현재 트래픽에 QoS를 제공할 수 없게 된다. 따라서 현재 클래스의 큐 길이의 최대값은 타겟 클래스의 하한보다 커야 한다.

Is the maximum queue length that can be set in class j as described above. The meaning of <Condition 1> is as follows. The length of the current traffic queue must fall within the boundary value of the inspection class. However, depending on the class, the interval of boundary values may be very narrow. In this case, the lower limit of the queue length of the inspection class may be larger than the average queue length of the current traffic. The queue length of the current traffic may not be within the range of the queue length of the inspection class. In this case, QoS can not be provided to the current traffic. Therefore, the maximum value of the queue length of the current class must be greater than the lower limit of the target class.

에서,

는 검사 클래스의 평균 큐 길이를 의미한다.

는 검사 클래스의 threshold 1을 의미한다. 검사 클래스의 평균 큐 길이가 검사 클래스의 threshold 1값보다 작아야 함은 당연하다. 만약 평균 큐 길이가 threshold 1값보다 크다면 안정적으로 QoS를 제공할 수 없기 때문이다. <Condition 2>

in,

Means the average queue length of the test class.

Means the threshold 1 of the test class. It is natural that the average queue length of the test class must be less than the threshold 1 value of the test class. If the average queue length is larger than threshold 1, QoS can not be reliably provided.

에서, 검사 클래스의 평균 큐의 길이가

가 threshold

보다 작아야 함을 의미한다.<Condition 3>

, The average queue length of the check class is

Is threshold

It should be smaller.

If the inspection class satisfies all of < condition 1 > to < condition 3 >, the inspection class becomes the target class. Since the target class has been found, the process of finding the target class for the next class is stopped.

의 경계값의 추정 방식에 대한 과정을 상세히 설명한다. 이하의 설명은 본 발명의 통계적 분석에 대한 수학적 배경에 해당한다. 이하의 설명들은 "W. Stallings, High-Speed Networks and Internets: Performance & Quality of Service, 2nd Ed. ISBN 0-13-032221-0, NJ: Prentice Hall, 2002. (Chapter 9)"에 기반한 것임을 밝혀둔다.Hereinafter, in order to facilitate the understanding of the present invention,

The method of estimating the boundary value of the boundary value will be described in detail. The following description corresponds to the mathematical background of the statistical analysis of the present invention. The following description is based on "W. Stallings, High-Speed Networks and Internets: Performance and Quality of Service, 2nd Ed. ISBN 0-13-032221-0, NJ: Prentice Hall, 2002. (Chapter 9) Leave.

The statistical values of the present invention are applied to the discrete class j traffic using Whitey's estimation method using a periodogram based on an intensity function. Thereby estimating the boundary value of the average queue length of class j traffic in real time (or near real time). For reference, periodogram is an estimate of the power spectral density of an irregular signal.

은 자기 유사 트래픽의 통계적 특성을 고려하였다. 클래스의 경계값을을 정밀히 구하기 위하여 필요한 수학적 도구는 자기상관함 수(autocorrelation function)와 파워스펙스럼밀도(power spectral density: PSD)이다. (One)

Considered the statistical characteristics of self - similar traffic. The mathematical tools needed to precisely determine the boundary values of classes are the autocorrelation function and the power spectral density (PSD).

The autocorrelation function can be expressed by Equation (12), and PSD can be expressed by Equation (13).

However, an additional process is required to estimate the autocorrelation function and PSD in real time (or near real time) using only limited sample data.

들을 기반으로 밀도함수(intensity function)를 계산할 수 있다.Analysis of self-similar traffic classified into multiple classes in real-time (or near real-time) is a discrete time instance value

The intensity function can be calculated based on the intensity function.

)는 하기 <수학식 14>로 표현된다.(1) density function (

) Is expressed by Equation (14) below.

(2) The calculation of the H (Hurst parameter) value and the sample variance value is calculated as follows using a periodogram.

을 최소값으로 만드는

값이 허스트 파라미터(Hurst parameter)의 추정값이 된다. 추정 값

을 이용하면 샘플 분산(sample variance)을 구할 수 있고 하기 <수학식 16>으로 표현된다. In Equation (15)

To a minimum

The value is an estimate of the Hurst parameter. Estimated value

The sample variance can be obtained and expressed by Equation (16).

Another modification of Equation (16) is Equation (17). The following expression (17) is a discrete value expression form.

와

가 추정되어 얻어지므로 바로 트래픽의 2차 모멘트(moment)

을 추정할 수 있다.In the equations (15) to (17)

Wow

The moment of the second moment of the traffic,

Can be estimated.

의 경계값을 계산하기 위한 다른 인자들 (즉,

)은 서버 시스템의 상태 정보와 스케쥴러로 입력되는 트래픽을 분석하여 실시간(또는 근 실시간)으로 계산할 수 있다. And

Lt; RTI ID = 0.0 &gt; (i. E., &Lt; / RTI &gt;

) Can analyze the state information of the server system and the traffic input to the scheduler and calculate it in real time (or near real time).

In addition, the number of packets arriving for the n-th time based on the fractional Brownian motion (FBM) based on the Norros technique (IEEE JSAC'95) can be estimated by the following Equation (18).

는 시간 구간을 나타낸다. 또한

는 허스트 파라미터(Hurst parameter)를 나타낸다. In Equation (18)

Represents a time interval. Also

Represents the Hurst parameter.

는 평균 입력율(mean input rate)을,

는 분산 계수(variance coefficient)를 의미하는데 상기 <수학식 18>에서

과

는 각각 하기 <수학식 19>와 <수학식 20>에 의하여 구해진다.Also

Is the mean input rate,

Is a variance coefficient. In Equation (18)

and

Are obtained by the following equations (19) and (20), respectively.

의 경계값을 추정하는 방식을 설명하였다. 이하에서는 본 발명에 따른 장치를 설명한다.By statistical estimation so far

The boundary value of the boundary is estimated. Hereinafter, an apparatus according to the present invention will be described.

5 is a block diagram of a scheduling apparatus for providing a differentiating service according to an embodiment of the present invention.

,

,

이다. The input interface 501 receives packets of traffic input to the scheduler, and delivers the packets to the controller. The variable value calculation unit 503 receives the information of the packets through the input interface 501. [ The information of the packets includes information on the number of packets arriving at the scheduler for a certain time (i.e., packet arrival rate). The variable value calculation unit 503 receives information of a server (not shown) connected to the output interface 511 through the output interface 511. [ The server information includes server capacity information. The variable value calculation unit 503 calculates the variable values used in the statistical analysis by using the packet reach rate and the server capacity information, and outputs the calculated values to the boundary value calculation unit 505. [ The values of the variables are calculated as described in the method of the present invention

,

,

to be.

,

)을 계산하여 비교부(507)로 출력한다. 비교부(509)는 입력인터페이스(501)로부터 현재 트래픽을 수신하고, 경계값계산부(505)로부터 경계값들을 수신한다. 또한 변수값계산부(503)로부터 변수값들을 수신한다. 비교부는 상기 수신한 현재 트래픽들, 경계값들, 변수값들을 이용하여 상기 현재 트래픽의 평균 큐 길이와 경계값들을 비교한다. 상기 비교 결과는 제어부(509)로 출력된다. 제어부는 상기 입력인터페이스(501)로부터 현재 트래픽들을 수신한다. 또한 상기 비교부(507)로부터 상기 비교 결과를 수신한다. 제어부는 상기 비교 결과가 전술된 클래스 이동 조건에 해당하는지를 검사한다. 클래스 이동 조건에 해당한다면 타겟 클래스를 검색한다. 타겟 클래스가 검색된 경우 상기 현재 트래픽의 클래스를 타겟 클래스로 변경하고, 상기 타겟 클래스가 검색되지 않은 경우 상기 현재 트래픽의 클래스를 최상위 클래스로 변경한다. 또한 제어부(509)는 상기 타겟 클래스로 변경된 현재 트래픽을 출력인터페이스(511)로 출력한다. 출력인터페이스(511)은 서버와 연결되어 있다. 서버는 상기 현재 트래픽을 처리하게 된다.The boundary value calculation unit 505 receives the variable values and outputs the boundary values (

,

And outputs it to the comparison unit 507. [ The comparison unit 509 receives the current traffic from the input interface 501 and receives boundary values from the boundary value calculation unit 505. [ And receives the variable values from the variable value calculation unit 503. The comparing unit compares the average queue length of the current traffic with the boundary values using the received current traffic, boundary values, and variable values. The comparison result is output to the controller 509. The control unit receives current traffic from the input interface 501. And receives the comparison result from the comparison unit 507. [ The control unit checks whether the comparison result corresponds to the above-described class moving condition. If the condition of class movement is satisfied, the target class is searched. If the target class is searched, the current traffic class is changed to the target class, and if the target class is not searched, the current traffic class is changed to the highest class. The control unit 509 also outputs the current traffic changed to the target class to the output interface 511. [ The output interface 511 is connected to the server. The server processes the current traffic.

FIG. 6 is a diagram illustrating an example of scheduling by the statistical estimation proposed in the present invention.

)가 3000 바이트(bytes)인 경우를 가정한 것이다. 표준편차(standard deviation)(

)가 1.5를 전후로 class 5가 class 4로 변경되었다. 또한

1.5에서 2사이에 class 4가 class 3으로 변경되었다.

가 2에서 2.5사이에 class 3에서 class 2로 변경되고,

가 3을 전후로 class 2가 class 1로 변경되는 것을 볼 수 있다. 이 같은 예시도에서 볼 수 있듯이, 본 발명에 따른 통계적 추정에 의한 스케쥴링에 의하여 현재 트래픽의 큐 길이가

= 3000 bytes를 초과하도록 클래스를 변경시키기 때문에 QoS를 보장하면서 트래픽이 처리되는 것을 볼 수 있다. FIG. 6 is a flow chart illustrating a method of setting a maximum queue length threshod (

) Is 3000 bytes (bytes). Standard deviation (

) Was changed to 1.5, and class 5 was changed to class 4. Also

Class 4 was changed to class 3 between 1.5 and 2.

Is changed from class 2 to class 2 from 2 to 2.5,

And class 2 is changed to class 1 before and after 3, respectively. As can be seen from the above example, the queue length of the current traffic is calculated by the statistical estimation scheduling according to the present invention

= 3000 bytes, it can be seen that the traffic is processed while ensuring QoS.

FIG. 1 is a block diagram of a general IP packet used for a differentiating service,

2 is a configuration diagram of a scheduling apparatus according to an embodiment of the present invention,

3 is an exemplary diagram for estimating an average queue length of traffic for each class of the differentiating service according to an embodiment of the present invention.

4 is a flowchart illustrating a scheduling method for providing a differentiated service according to an embodiment of the present invention;

5 is a configuration diagram of a scheduling apparatus for providing a differentiating service according to an embodiment of the present invention;

6 is a diagram illustrating an example of scheduling by statistical estimation proposed in the present invention.

Claims (12)

A scheduling method of a differentiated service system, Checking whether the current traffic is self-similar traffic, Checking whether the current traffic satisfies a class moving condition when the current traffic is the self similar traffic; Searching for a target class satisfying quality of service (QoS) of the current traffic when the class moving condition is satisfied; When the target class is searched, changing a source class to which the current traffic belongs to the target class, and if the target class is not searched, changing the source class to a highest class among a plurality of predetermined classes The scheduling method comprising the steps of: The method according to claim 1, The average queue length of the primitive class may be, Wherein the scheduling information is determined in consideration of the arrival rate of the current traffic per hour, the total capacity of the server allocated to the current traffic, and the remaining capacity of the server. 3. The method of claim 2, The step of checking whether the class moving condition is satisfied, Using the one of the proximity of the upper limit value of the average queue length and the proximity of the maximum value of the allowable queue length to the average queue length. The method of claim 3, The proximity to the upper limit value and the proximity to the maximum value may be determined, Wherein the determination is made using an argument value that varies depending on an operation mode of the system, a traffic characteristic, and a class priority. The method according to claim 1, And increasing a processing capacity of a server allocated to the current traffic while maintaining the original class when the target class is searched. delete An apparatus for scheduling a differentiated service system, Checking whether the current traffic is a self-similar traffic, checking whether the current traffic satisfies the class moving condition when the current traffic is the self-similar traffic, checking if the current traffic satisfies the class moving condition, (QoS), and when the target class is searched, changes the source class to which the current traffic belongs to the target class, and if the target class is not searched, And a control unit for changing to a highest class among a plurality of determined classes. 8. The method of claim 7, Further comprising a comparison unit for determining an average queue length of the primitive class in consideration of the arrival rate of the current traffic, the total capacity of the server allocated to the current traffic, and the remaining capacity of the server. 9. The method of claim 8, Determining a proximity of an upper limit value to the average queue length and a proximity of a maximum value of a queue length allowable to the current traffic with respect to the average queue length to calculate whether or not the class moving condition is satisfied; And a scheduling unit. 10. The method of claim 9, The boundary value calculation unit may calculate, Wherein the proximity to the upper limit value and the proximity to the maximum value are determined using a parameter value changed according to an operation mode of the system, a traffic characteristic, and a class priority, respectively. 8. The method of claim 7, Wherein, And when the target class is searched, increasing the processing capacity of the server allocated to the current traffic while maintaining the original class. delete

Images