CC BY
66
TRAFFIC FLOW MANAGEMENT MODEL IN SOFTWARE-DEFINED NETWORKS WITH UNEQUAL LOAD METRIC
Krasov
Andrey Vladimirovich,
Ph. D., docent, chief of the SPSUT department of Secured Communication, St. Petersburg State University of Telecommuncations, St. Petersburg, Russian Federation, krasov@pisem. net
Levin
Mark Vadimovich,
Assistant on the SPSUT department of Secured Communication, St. Petersburg State University of Telecommunications,
St. Petersburg, Russian Federation, m. va. levin@gmail. com
Shterenberg Stanislav Igorevich,
Assistant on the SPSUT department of Secured Communication, St. Petersburg State University of Telecommunications,
St. Petersburg, Russian Federation, [email protected]
Isachenkov Pavel Andreevich,
Master's student on the SPSUT department of Secured Communication, St. Petersburg State University of Telecommunications,
St. Petersburg, Russian Federation, [email protected]
O
Keywords:
software-defined networks; load; traffic management; resiliency; availability; metre; routing.
In this work the traffic management model, based on the telecommunication load is studied. This method can be applied to the IP-based software defined networks. EIGRP routing protocol algorithms were studied as well. The dependence of EIGRP path metric on changing load in modeled IP network was plotted using MatLab. Found results showed that the default behavior of EIGRP protocol in relation to the calculated metrics with the load taken into account, in the event that this option is activated, is unstable and probabilistic in nature and leads to a state where the EIGRP route metrics are constantly recalculated, resulting in a random rerouting and loss of control over the management of IP network, which in turn leads to a denial of service and disruption of an availability of information, which is one of the main elements of information security. To prevent such behavior it is suggested to use modified EIGRP routing model - to carry out on to the separate device, the controller, a method of traffic management, which on a centralized basis collects information on the active load on the device interfaces. To process the information obtained from the devices, controller uses the algorithm which in turn allows considering not absolute, but relative change of load on the interfaces on the network devices. Within the described algorithm it is offered to used different values of coefficients, which, when changed, can allow bigger or smaller sensitivity of the offered algorithm to the relative changes of the load on interfaces of network devices. Using numerical modeling the plot of the dependence of path metric, calculated by EIGRP, on changing load in the modeled IP network, with the described algorithm taken into consideration, was built. The results showed that using the described method can stabilize the work of EIGRP and provide more control over the IP network, which in result can prevent failures in service and provide the information accessibility.
Introduction
In modern data transfer networks operating on the TCP/IP protocol stack, one of the most commonly used routing protocols is EIGRP or Enhanced Interior Gateway Routing Protocol. At first this protocol used to be proprietary and it was owned by its developers, Cisco Systems, who opened their creation to the general public as an Informational RFC, allowing other vendors to implement the protocol in their devices.
Initially, EIGRP was presented as a "hybrid" routing protocol that combines features of both link-state and distance-vector protocols. In fact, it is more correct to call it an enhanced distance-vector routing protocol. Unlike other protocols based on that algorithm, which are waiting for the route information updates while in dormant state, EIGRP inquires it by itself.
Unlike its predecessor, IGRP or Interior Gateway Routing Protocol, which was also designed by Cisco Systems, EIGRP does not rely to Bellman-Ford algorithm for finding the shortest path, but instead it uses Diffusing Update Algorithm or DUAL. There are some terms that can be associated with this algorithm:
• Successor - a neighboring router, through which the packets can be sent with the minimal cost;
• Advertised Distance (AD) - the cost (or the metric) of a route, which comes through the closest router to the destination network;
• Feasible Distance (FD) - AD value summarized with the cost of the route from closest router to the local one;
• Feasible Successor - a spare router, through which there is a route without loops;
• Feasible Condition - a condition stating if a router can become a Feasible Successor. (AD of a possible next router should be less than the FD of current route).
DUAL operates on the principle of a Finite State Machine (FSM). This way the work process for this algorithm is that if the connection with a successor is interrupted, the FSM will check the network for a possible successor, and, if there is one, it will become primary. After that the successor will be added to the routing table. Possible successors (if there are some) will be added to the topology table as well. If there are no Feasible Successors, the FSM puts the destination network into active state and requests new possible routes from its neighbours. In case where there is a possible neighbour, the successor is added to the routing table and all actions for the feasible successor are repeated. If there is no possible new route to the destination network, it gets deleted from topology and routing tables.
There are three tables used in EIGRP:
• Neighbor table - contains information about routers connected to a current one;
• Topology table - contains information about routes;
• Routing table - used while there has to be a decision about forwarding packets.
Implementation
To evaluate a quality of a route a metric is used. Metric is a number that represents some characteristics of a route or their aggregate. EIGRP has 5 of those characteristics, while only 2 are used initially:
• Bandwidth - lowest aggregate bandwidth between the source and the destination;
• Delay - cumulative delay of all of the interfaces on the route;
• Reliability - worst reliability on the route;
• Load - worst load on the route;
• MTU - lowest MTU on the route. Is not used to calculate the metric, but still included into the EIGRP updates.
To calculate the metric K1,K2,K3,K4,K5 coefficients are used. By default, Ki=K2 = 1, K2 = K4 = K5 = 0. Values of given coefficients can vary from 0 to 255.
The formula to calculate the EIGRP composite metric is as follows:
M = (K1*BW + (2Sg62^J + K3 * Delay) *
(-^-) * 256 (1)
Kreliability+K^J v 7
Where:
BW = 256 *-
BWmin(Kbps) '
Delay = 256 * (ïï^).
(2) (3)
Where Delayn - is an aggregate of all of the delay values on all of the interfaces.
With constant load changes, recalculation of metrics with this formula can lead to a random route change and to loss of control of data network. This outcome leads to a denial of service, which can not be a desirable situation.
To prevent such behavior, it is proposed to apply an algorithm, which takes into consideration previous load values. To investigate the effect, this algorithm will have on the route metric, a simple network model was created (Fig. 1).
Fig. 1. Researched network topology. In purposes of the research, the following values were cho-
BW1 = 100 Mbps = 100000 Kbps; BW2 = 8 Mbps = 8000 Kbps; Del1 = 100 msec = 0.0001 sec; Del2 = 5000 msec = 0.005 sec; Rel = 255.
As a Load, values of a uniform distribution function are chosen.
Values of a coefficients, used in the research:
• Kt = K2 = K3 = 1;
• K4 = K5 = 0.
Thus the formula (1) takes the form of:
см
f256*10^4 r£îh + Del2)))
( m ЛП )))
BW2 ) 256-Load.
+ К3 * (256 *
(4)
Without taking into consideration the previous load values, the metric will appear as shown in the figure (fig. 2).
Fig. 2. Metric values depending on time.
As it can be seen on the graph, at some moments the value of metric significantly increases, creating a "peak". This may adversely affect the operation of equipment.
The main problem of EIGRP, when working with load parameters, is how unstable the routes are, because of the probabilistic nature of the load. As a result, current protocol implementations do not use these parameters to calculate the metric. To get around these limitations, it is possible to use an adaptive algorithm, which responds to the load changes [3,4] and defined by difference equation:
Loadi
a * Loadi_1 + (1 — a) * Load„ 0 <
(5)
a
lnew> <1
The calculations are done for those router interfaces, which parameters controller was able to get.
Fig. 3. Algorithm working process in the software-defined network
After the threshold value of a parameter (given by an administrator) was reached due to changes in load values, the controller transmits to a router (or several routers) the decision to recalculate the route. Since the router participating in the EIGRP working process know the current load values, the transmission of those parameters is not required. After receiving the decision whether to recalculate the routes, the router begins metric calculations either for all of the routes or only for those, which are affected by changing metric values according to the current EIGRP specifications without any modifications. This kind of final realization of proposed mechanism uses already existing on the network devices mechanisms and algorithms and it does not require any kind of changes neither in the hardware nor the software of network devices [3,4]. The proposed algorithm with all of the calculations can be deployed on a controller, which can be any platform [2], software or hardware, with the support of the relevant APIs.
As a result of the implementation of an adaptive algorithm, the EIGRP metric takes the form shown in the figure below (fig. 4):
According to the traits of differential equation, the value of the Load parameter, calculated for the current time period on the controller will depend on the values of Load, calculated on the previous time period and the values of Load parameter, acquired from the router for the current time period. With this equation the value of the latter in the total calculation result will depend on the value of coefficient a. While this coefficient is increasing, the sensitivity of a current algorithm to changes in the load is decreasing and vice versa. The question of how to choose the value of the coefficient in case of specific topologies and traffic patterns (load probability distribution law) is open and requires further investigation.
At the controller level [1,5] the threshold value is set, and it provides the condition of reaction on changes in load. With the setting of an a coefficient, it determines the complete reaction of an algorithm to load changes in the network.
Since for the calculation of metrics for the route, the router uses integer numerical values of all the parameters (bandwidth, delay, load and reliability), the controller-based calculations should be rounded to the nearest integer value, to subsequently pass those to the router.
Fig. 4. Metric values depending on time, after the implementation of the adaptive algorithm
As it can be seen from the plot (fig. 4) the metric values are changing a lot smoother and there are almost no drastic changes in values (peaks).
On the image above (fig. 5) metric values before and after the implementation of the algorithm are together. The dotted line is the values with the standard working process of EIGRP, while
the regular line is what the metric values are after the implementation of proposed algorithm. As it can be seen from the plot, the reaction to the significant change in the metric is not that substantial. This can help set the certain threshold value of that change, for which it will be not necessary to recalculate the route.
Fig. 5. Metric values before and after implementation of the adaptive algorithm
The control process of a LAN based on the short-term prediction of the spread of self-modifying code across the network nodes can be presented as a closed loop, consisting of separate phases (fig. 6). The first four phases define the cycle of processing and analyzing the information, while the rest define the control cycle.
Information processing and analyzing cycle solves the problems of generalization, processing and defining of the status of the self-modifying code (SMC) in the network nodes at a certain time; the problems of transmission of the given information to the network control center (NCC). The control cycle is responsible for prediction of the SMC spread through the network nodes, as well as for the decision-making in case of acquired data.
Information processing and Eimlyzmg cyde
F
'orming the information about
the SMC state
Transferring the state information
Generalization and analyzis of the SMC spread state
NCC control program
I
dentification of a state
SMC s
Control cycle
Prediction of a SMC spread state through the network nodes
Forming controlling actions 1
Decision-making about returning the required network security level
Getting the control commands to the controlled object
Execution of those commands 1
Control of the command execution
Fig. 6. Control process of the LAN NCC
In order to predict the spread of SMC, it was decided to use one of the control methodologies based on the Model Predictive Control (MPC). This methodology is already in use for almost half of a century in such areas like oil refining industry, medicine, energy, robotics, etc., which cans show
that this methodology is effective and universal, as well as time-proved.
The development process for the control method for LANs based on short-term prediction of the SMC spread can be divided into the following stages: a description of a SMC spread model, creating some criteria for the safe state of a LAN, the calculation of a control rule.
One of the advantages of the control methodology using the prediction models is the possibility of study multifactorial process in advanced mode. Therefore, the prediction method based on the state-space model (SSMPC) is selected. To apply this method, it is required to create a mathematical model of the controlled object, which is later used in the prediction of an output data of a LAN on the basis of the past and current values and the estimated optimal controlling impacts in future. These impacts are calculated by an optimizer which also takes the quality criteria and the restrictions, imposed to the process variables, into consideration.
The selected model should encompass the dynamics of a process for an accurate prediction of future output values, it should be simple and easy to implement and understand.
Conclusion
As a result of this study, using the numerical modelling in MATLAB application, the plots of dependence of EIGRP metric on the changing load values in the IP data network were obtained. Those plots showed the difference between the standard EIGRP working process and the one with using of proposed algorithm. The results showed that using described method allows, firstly, to stabilize the EIGRP work, and, secondly, to provide greater control over an IP data network, which in combination, helps prevent the denial of service and provides the accessibility of information.
References
1. Thomas D. Nadeau, Gray K. SDN: Software Defined Net-works. Sebastopol: O'Reilly, 2013. 350 p.
2. Azodolmolky S. Software Defined Networking with Open-Flow. Birmingham: Packt Publishing, 2014. 152 p.
3. Krasov A. V., Levin M. V. Opportunities management traffic of a concept within SDN. IV International scientific-technical and scientific-methodical conference: collection of scientific articles in 2 volumes. Actual infotelecommunications problems in science and education, St. Petersburg, 03-04 March 2015 St. Petersburg, St. Petersburg State University telecommunication them. prof. M.A. Bonch-Bruevich, 2015. Pp 350-354.
4. Krasov A. V., Levin M.V., Tsvetkov A.Y. Management data networks with varying load // All-Russian scientific conference on the problems of the Power Management-Technical Systems, St. Petersburg, 28-30 October 2015 St. Petersburg, St. Petersburg State Electrotechnical University, 2015. No. 1. Pp. 141-146.
5. Chugreev D.A., Shkrebets A.E., Shevel A.E., Vlasov D.V., Grudinin V.A., Kairkanov A.B., Gardens O.L, Titov V.B., Horuzhnikov S.E., Soames LN Software-configurable network: OpenFlow and virtual network overlap // Modern problems of science and education. 2013. No. 4. Pp. 55.
For citation:
Krasov A.V., Levin M.V., Shterenberg S.I., Isachenkov P.A. Traffic flow management model in software-defined networks with unequal load metric. H&ES Research. 2016. Vol. 8. No. 4. Pp. 70-74.
МОДЕЛЬ УПРАВЛЕНИЯ ПОТОКАМИ ТРАФИКА В ПРОГРАММНО-ОПРЕДЕЛЯЕМОЙ СЕТИ С ИЗМЕНЯЮЩЕЙСЯ НАГРУЗКОЙ
Красов Андрей Владимирович,
Санкт-Петербург, Россия, [email protected]
Левин Марк Видимович,
Санкт-Петербург, Россия, [email protected]
Штеренберг Стнислав Игоревич,
Санкт-Петербург, Россия, [email protected]
Исаченков Павел Андреевич,
Санкт-Петербург, Россия, [email protected]
Аннотация
Рассматривается модель управления трафиком, основанный на учете телекоммуникационной нагрузки, который возможно применить в программно-определяемых сетях передачи данных на основе 1Р. Исследованы алгоритмы работы протокола маршрутизации БЮКР. Путем численного моделирования в прикладном пакете MatLab получены графики зависимости метрики маршрута, вычисляемой БЮКР, от изменяющейся нагрузки в моделируемой 1Р-сети передачи данных. На основе полученных результатов показано, что стандартное поведение БЮКР в отношении вычисляемой метрики при учете нагрузки, в том случае, если такая возможность активирована, носит нестабильный вероятностный характер и при-водит к состоянию, при кото-ром метрики маршрутов БЮКР постоянно пересчитываются, что приводит к случайному изменению маршрутов и потере контроля управления над 1Р-сетью передачи данных, что, в свою очередь, приводит к отказам в обслуживании и нарушению свойства доступности информации, как одного из элементов информационной безопасности. Предложено, для предотвращения такого поведения, использовать модифицированную модель маршрутизации БЮКР - вынести на отдельное управляющее устройство - контроллер - метод управления трафиком,
который централизованно собирает со всех сетевых устройств информацию о действующей на интерфейсах этих устройств нагрузке; контроллер, для обработки полученной от сетевых устройств информации, использует алгоритм, который, в свою очередь, позволяет учитывать не абсолютное, а относительное изменение нагрузки на интерфейсах сетевых устройств. Предложено также, в рамках описанного алгоритма, использовать различные значения коэффициентов, изменения которых позволяют обеспечить большую или меньшую чувствительность предложенного алгоритма к относительным изменения нагрузки на интерфейсах сетевых устройств. Путем численного моделирования в прикладном пакете MatLab получены графики зависимости метрики маршрута, вычисляемой БЮКР, от изменяющейся нагрузки в моделируемой 1Р-сети передачи данных с использованием описанного в статье метода управления трафиком, показывающие, что применение описанного в статье метода, позволяет, во-первых, стабилизировать работу БЮКР, во-вторых, обеспечить больший управляющий контроль над 1Р-сетью пере-дачи данных, что, в совокупности, позволяет предотвратить отказы в обслуживании и обеспечить свойство доступность информации.
Ключевые слова: программно-определяемые сети; нагрузка; управление трафиком; отказоустойчивость; доступность; метрика; маршрутизация.
Информация об авторах:
Красов А.В., к.т.н., доцент, зав. каф. Защищенных систем связи, Санкт-Петербургский университет телекоммуникаций им. проф. М.А. Бонч-Бруевича;
Левин М.В., ассистент каф. Защищенных систем связи, Санкт-Петербургский университет телекоммуникаций им. проф. М.А. Бонч-Бруевича;
Штеренберг С.И., ассистент каф. Защищенных систем связи, Санкт-Петербургский университет телекоммуникаций им. проф. М.А. Бонч-Бруевича; Исаченков П.И., студент-магистр каф. Защищенных систем связи, Санкт-Петербургский университет телекоммуникаций им. проф. М.А. Бонч-Бруевича.
Для цитирования:
Красов А.В., Левин М.В., Штеренберг С.И., Исаченков П.А. Модель управления потоками трафика в программно-определяемой сети с изменяющейся нагрузкой // Наукоемкие технологии в космических исследованиях Земли. 2016. Т. 8. № 4. С. 70-74.
