Fuzzy Logic Tsukamoto for SARIMA On Automation of Bandwidth Allocation

The wireless network is used in different fields to enhance information transfer between remote areas. In the education area, it can support knowledge transfer among academic member including lecturers, students, and staffs. In order to achieve this purpose, the wireless network is supposed to be well managed to accommodate all users. Department of Electrical Engineering and Information Technology UGM sets wireless network for its daily campus activity manually and monitor data traffic at a time then share it to the user. Thus, it makes bandwidth sharing becomes less effective. This study, build a dynamic bandwidth allocation management system which automatically determines bandwidth allocation based on the prediction of future bandwidth using by implementing Seasonal Autoregressive Integrated Moving Average (SARIMA) with the addition of outlier detection since the result more accurate. Moreover, the determination of fixed bandwidth allocation was done using Fuzzy Logic with Tsukamoto Inference Method. The results demonstrate that bandwidth allocations can be classified into 3 fuzzy classes from quantitative forecasting results. Furthermore, manual and automatic bandwidth allocation was compared. The result on manual allocation MAPE was 70,76% with average false positive value 56 MB, compared to dynamic allocation using Fuzzy Logic and SARIMA which has MAPE 38,9% and average false positive value around 13,84 MB. In conclusion, the dynamic allocation was more effective in bandwidth allocation than manual allocation. Keywords—Bandwidth allocation management; dynamic allocation; fuzzy logic; Tsukamoto inference method; SARIMA


I. INTRODUCTION
Computer network becomes the important aspect in data communication.Thousands of user are using the network to transfer information to the remote area.The wireless network is common of computer network technology that is widely used in many different institutions to achieve different activity, including educational institution [1].Therefore, network bandwidth in campus is supposed to be managed to meet user needs.
Bandwidth is channel capacity or the maximum throughput of a physical or logical communication path in a digital communication system [2].The higher bandwidth consumption needs a good management.Bandwidth management is a way to achieve optimum usage with limited available bandwidth [1].
Universitas Gadjah Mada (UGM) as an educational institution is also implementing Wireless Local Area Network (LAN) or Wi-Fi on their campus.This Wi-Fi is installed at all faculties in UGM.Every UGM staffs and students are provided with their own username and password to connect to the internet through the Wi-Fi.The number of users who connects to access point tends to rise.Thus, it is necessary to manage bandwidth efficiently.
The earlier study, especially in Electrical Engineering and Information Technology presented that bandwidth usage had a seasonal pattern.Seasonal Autoregressive and Moving Average (SARIMA) was applied to predict bandwidth usage.The results showed that bandwidth traffic tends to rise on Tuesday and get down on Sunday.Traffic fluctuations indicate the weekly bandwidth allocation should be changed [3].

Indra Hidayatulloh et al, added Exponential Generalized
Autoregressive Conditional Heteroscedasticity (EGARCH) to reduce heteroscedasticity effect on bandwidth prediction using SARIMA.SARIMA-EGARCH increased prediction accuracy around 19,15% compared to stand alone SARIMA.Unfortunately, the auto-correlation effect happened during prediction on time series bandwidth data [4].
One of the methods of bandwidth management is by using system scheduling [5].Mikrotik outerOS™, that is applied at UGM Hotspot, is a network router based on Linux.Mikrotik is also supported by Windows application (WinBox) to ease its router adjustment.Some scheduling methods are implemented at Mikrotik RouterOS, for example, Simple Queue, Per Connection Queue (PCQ) and Hierarchical Token Bucket (HTB).Simple Queue is used to restrict the number of data for specific addresses or subnet [5].It is the simplest scheduling since the limitation of maximum upload and download is implemented refer to IP address client.PCQ method can divide bandwidth automatically from the active users.Additionally, this method also has disadvantages, whereas may result in bandwidth leakage or unfair division [6].HTB method implements link sharing so that the residual bandwidth in a class node can be distributed into another class.It uses Token www.ijacsa.thesai.orgBucket Filter (TBF) as an estimator in the determination of bandwidth allocation.It is more adaptive since TBF provides bandwidth in rate.In HTB method, the admin needs to define the maximum rate in every class node.Unfortunately, it is still doing manually based on admin intuition.Therefore, the bandwidth management is less efficient and less adaptive.
Refer to the problem, there is important to develop an adaptive model for bandwidth management.A model must be able to provide the values of maximum rate automatically.The forecasting results could not be used directly for system input because it has a decimal data type with wide class boundaries that will be hard for a network administrator to manage these quantitative values.Moreover, forecasting results still have more error potential if the data applied directly rather than using a specific range.These data have to be converted into classification form to determine the maximum rate allocation.Fuzzy Logic was implemented to achieve this goal.Some studies on bandwidth management have investigated.A previous work [7] has explored rate control strategies for real-time multimedia variable bit rate (VBR) services.It was implemented in IEEE 802.16 broadband wireless networks.This study managed bandwidth allocation on max-min fairness queue scheduling using a time constraint condition.Liu, et al [8] has predicted network traffic by using chaos theory and Support Vector Machine (SVM).This research used campus data including wired and wireless.The forecast values could be used to manage the bandwidth.A proposed scheme dynamically reserves and allocates bandwidth based neural network has been studied by Song et al. [9].It was applied to different types of calls.Lee et al [10] has implemented roundrobin schedule to allocate bandwidth.Prediction of Available Bandwidth Estimation with Mobility Management in Ad Hoc Networks has been undertaken by Belbachir et al. [11].Hierarchical game theory models were also be implemented for bandwidth management [12].While Maestrelli et al. [13] proposed quantization model for bandwidth adjustment.
The aim of this research was to develop a Fuzzy Logic Tsukamoto in order to support bandwidth allocation decision automation system called BIOMA (Bandwidth Automation Management).Fuzzy Logic Tsukamoto uses monotone membership function with Center Average Deffuzzyfier method.By this defuzzification method, Tsukamoto selects mean from the range given.Meanwhile, Mamdani method selects Minimum or Maximum Value.If the minimum is selected, it might affect bandwidth allocation doesn't meet the network requirement.In the other side, if the maximum is selected, it might cause bandwidth being extravagant.The other method, Sugeno Method, gives consequences value as crisp values using some linear calculation.So this method doesn't meet the system requirement like Fuzzy Logic Tsukamoto.
The remainder of the paper is structured as follows.Section 2 illustrates data input.Section 3 presents the methodology.Section 4 describes experimental results.Finally, Section 5 presents the conclusion of the study.

II. DATA INPUT
Forecasting results of bandwidth usage were used as input data [2].It was used real downstream dataset (Mbps).Data was collected from monitoring Universitas Gadjah Mada (UGM) portal at http://mon.ugm.ac.id/cacti/weathermap.This study collected time series data for the 20 week period from 09 September 2013 to 27 January 2014.The original data is plotted as presented in Fig. 1.Bandwidth usages were predicted by using SARIMA method with outlier detection.Refer to previous research [2], the most appropriate SARIMA model was (0,1,1)(0,1,1)7C from various traces.The observation found that some outliers in data collection influenced forecast accuracy.This problem was solved by including outliers detection to the model.Missing value analysis was done by using mean substitution operation to handle outliers.The approach could reduce forecasting error (MAPE) into 14.49%.
From computation, the parameter results were MA(1) = 0.9519 and SMA(7) = 0.9246 and constant = 0.010686.The estimated parameters were included to form the final model that expressed as backshift model shown in equation (1).

III. METHODOLOGY
In order to build adaptive and dynamic bandwidth management, it is necessary to build a system that can model bandwidth needs and give maximum value rate automatically.To do this, SARIMA method was used to predict bandwidth needs and Fuzzy Logic method was used to allocate bandwidth dynamically.

A. SARIMA Model Transformation
The SARIMA model for this research is model (1) which transformed into regular equation form in order to be used as system model.
The model used in the research as follows.

B. Fuzzy Logic Method
There is some approach that can be implemented to bandwidth allocation.One of them is a heuristic model.Fuzzy Inference System (FIS) is a heuristic model that widely used [6].This study implemented Fuzzy Logic Tsukamoto.This method was represented by a fuzzy set with a monotonical membership function.The monotonical reasoning was used when two Fuzzy areas are related with the following implication: (6)

𝑦 𝑓 (𝑥 𝐴) 𝐵
The implication function extends monotonical reasoning into: For example, there are two input variable Var-1(x) and Var-2(x), and output variable Var-3(z).Var-1 is divided into two sets:  and  .Var-2 is divided into set  and  .Whilst, Var-3 is divided into sets:  and  , whereas  and  are supposed to be monoton.Therefore, two rules are used: The first step is finding membership function for each fuzzy set in its rule.Sets of  ,  , and  are driven from fuzzy rule [ ] and sets  ,  , and  come from fuzzy rule [ ]. Fuzzy rule and may be represented in determining crisp values Z. Furthermore, the inferred output for each rule is defined as a crisp value induced by the rule's firing strength (α-predicate).www.ijacsa.thesai.org The final output result is taken from the weighted average of the output of each rule [15].Fig. 2 illustrates steps of Tsukamoto method.

IV. EXPERIMENTAL RESULT
This following section is discussing the result of Tsukamoto steps in order to manage bandwidth allocation on available dataset.

A. Construction of Membership Function
This step is focus on the developing a fuzzy set.There were 2 variables for modeling, "Usage" is for input variable and "Allocation" is for output variable.The Usage has 3 fuzzy sets, they are Small, Medium, and Large."Usage" is real type variable with its own domain:

B. Identify the Headings Rule Formation
Rules are one of FIS requirement [16].The calculation includes Usage as input variable and Allocation as output variable.It yields the following rule format:

D. Allocation Calculation
The calculation was started with the determination of membership weight.Membership weight for each data was calculated based on 2 domain, with each domain consists of Small, Medium, and Large.The result is presented in Table 2.  Where, x are time series from SARIMA forecasting results.
Next step was determining y values that represented as allocation values for each time series by using formula (14).The calculation results are shown in Table 3. V. BIOMA SYSTEM IMPLEMENTATION Bioma system has been implemented with a user friendly interface design using Bootstrap.Fig. 4 shows the interface of dashboard page that contains history of bandwidth allocation and bandwidth prediction the next few days.Bioma system tested by comparing MAPE between manual/ static and dynamic allocation using Bioma.The test results are shown in Table 4 below:

Fig. 3 .
Fig. 3. Triangle curve of data.Membership function for each set has been calculated as the following:

[
]                (11) where:  : fuzzy rule-i (i=1..m).  : weight values of Usage-i   : fuzzy set of Himpunan weight values of Usage-i ¤ : operator n : number of data   : fuzzy set for allocation variableC.Weight Calculation and Determination of Bandwidth AllocationThe previous rules were used to determine each data weight.Values of predicate α were found from rule composition (  ).The predicates are associated differently with the operator.In AND operator, predicate value of "  is  1   2   2 " is given as:      (12)      (13)One consequent values are obtained, y values can be calculated as: www.ijacsa.thesai.org

Fig. 5
Fig. 5 as follows is the interface of schedules page that used to set bandwidth allocation on specific date.
Table 1 is a month prediction of bandwidth usage in the Department of Electrical Engineering and Information Technology, UGM.

TABLE I
www.ijacsa.thesai.org [14]equently, based on the same book[14]a model that has an MA time series character, in both seasonal and regular part, will require one or more multiplicative terms to be combined on the model.Therefore, final model of   as follows.By combining the formulas (4-2) and (4-3) the value of the predicted value   can be determined as follows.   1   7   8  1   1  1 7   7

TABLE III .
RESULTS OF BANDWIDTH ALLOCATION