System Architecture: Dynamic Decision Estimation Module With QoS Control

Abstract and 1 Introduction 2 Related Work 3 Image Compression Module 4 System Architecture 5 Experimental Results 6 Conclusions A Compression Results References 4 System Architecture The proposed system provides an efficient interactive streaming desktop remote service for diversified client devices and dynamic network environments. The overall system structure is shown in Figure 4. Whenever a device requests a multimedia desktop streaming service, by the agent is installed on the client device, it helps transmit data included hardware and network environment parameters to User Profile Module (UPM). After that the UPM records the device codes and determines the required parameters, and then transmits them to the Dynamic Decision Estimated Module (DDEM). By analysis of historical data, the DDEM determines the most suitable compression ratio µ number for the device according to the parameters and the network condition. The µ number is transferred to the Image Compression Module (ICM) for use in the encode phase. Finally, the multimedia video file is transmitted to the mobile device through the service. 4.1 User Profile Module The profile agent is used to receive the device hardware environment parameters and create a user profile. The device transmits its hardware specifications in XML-schema format to the profile agent in the cloud server, as shown in the Table 1. The XML-schema is meta-data, which is mainly semantic and assists in describing the data format of the file. The meta-data enables non-owner users to see information about the files, and its structure is extensible. However, any device that is using this cloud service for the first time will be unable to provide such a profile, so there shall be an additional profile examination to provide the test performance of the mobile device and sample relevant information. Through this function, the client device can generate an XML-schema profile and transmit it to the profile agent. The profile agent determines the required parameters for the XML-schema and creates a user profile, and then transmits the profile to the DDEM for identification. 4.2 Dynamic Decision Estimation Module with QoS Control Figure 5 illustrates the DDEM used to estimate the value of σ. Specifically, the DDEM conducts this estimation after receiving the profile data from the User Profile Module. Based on multiple linear regression model in statistics, Based on multiple linear regression model in statistics, the DDEM determines the relationship between factors and predicts a suitable range for the number µ number. The problem is how to choose the best µ to obtain a suitable image size after the compression phase while also ensuring sufficiently high image quality and satisfying the image resolution demanded by the user. The ICM is used to obtain the compression data and calculate the corresponding P SNR values; these data are defined as historical data for the simulations, as shown in Table 3. In the current application, PSNRs of 25 to 50 dB are considered to be acceptable in terms of transmission quality. Therefore, before the regression analysis, we need to extract a list of values filtered by the condition P SNR ≥ 25. After we have a suitable range for the µ ratio, the system will analyze which µ number is most suitable for the compression phase. Specifically, let σ be the available bandwidth and θ the bandwidth threshold. If σ is greater than θ, the system can transmit the video content with the best possible quality without further processing. Otherwise, the next step is to determine appropriate values of P SNR and Compression Size to adapt to the current network conditions. In this case, the P SNR and Compression Size values are looked up in the estimation table; this table comprises a set of tuples of the form h(Resolution, P SNR, Compression Size), as described in the next section. Eventually, these tuples obtained is input to the estimate model, then the output µ value is asymptotic as well. The operation of QoS control is described algorithmically as Algorithm 3. 4.2.1 Proposed Estimation Model In the present paper, we use multiple linear regression (MLR) to estimate the value of a variable based on the value of two or more other variables. The goal of MLR is to model the statistical relationship between response and predictor variables. A simple linear regression illustrates the relation between the dependent variable y and the independent variable x based on the regression equation According to the multiple linear regression model the dependent variable is related to two or more independent variables. The general model for k variables is of the form To simplify the computation, the multiple regression model in terms of the observations can be written using matrix notation. Using matrices allows for a more compact framework in terms of vectors representing the observations, levels of regression variables, regression coefficients and random errors. The model is in the form and when written in matrix notation we have where Y , X denote respectively the response values for all observations into a n-dimensional vector and predictor into a n×p+ 1 matrix. β denote the intercepts and slopes into a k + 1-dimensional vector and is an n × 1 vector of random errors. The first step in multiple linear regression analysis is to determine the vector of least squares estimators βb which gives the linear combination yb that minimizes the length of the error vector. Basically the estimator βb provides the least possible value to sum of the squares difference between yb and y. Now, since the objective of multiple regression is to minimize the sum of the squared errors, the regression coefficients that meet this condition are determined by solving the least squares normal equation In the present work, µ is considered to be the response variable and the Compression Size, Resolution and P SNR are used as predictor variables. In this case, we defined the Resolution as a value from 1 to 7 corresponding to one of seven common screen resolutions; these resolutions are listed in Table 3. Thus, from Eq. 3, we have the following linear model for this approach. In linear regression, one problem with least squares occurs when there are one or more large deviations, i.e. cases whose values differ substantially from the other observations. These points are called outliers. There are various methods for outlier detection [54]. As shown in Figure 6, using Cook’s distance which is commonly used of the influence of a data point when performing least squares regression analysis. Cook’s distance used to indicate data points that are particularly worth checking for validity. In this case, there are several outliers above the reference line; these are identified and removed from the model. After reducing the outliers, we obtain the following fitted regression model that represents the data adequately. The difference between the observed and predicted score, Y-Y’, is called a residual. We analyze a histogram that shows the range of the residuals and their frequencies, and a probability plot that shows how the distribution of the residuals compares to a normal distribution with matching variance. As shown in Figure 7, the new residuals plot looks fairly symmetric and the probability plot seems reasonably straight, suggesting that the data fits reasonably to normally distributed residuals. 4.2.2 Estimated Table Specifically, the system will automatically adjust to increase or decrease the PSNR according to the available bandwidth σ representing the current network status. Then, by looking up the predictor variables P SNR, Resolution and Estimated Size in the estimated table, we can find the best tuple to input into the estimate model. Obviously, the µ value obtained here satisfies the client demands. For example, consider a case in which a thin client device sends a request to the system, and the available bandwidth σ is less than or equal to a predefined threshold value θ representing 80 `o the full bandwidth. The system will then decrease the PSNR value. The default is 30 dB. Assume we would decrease the PSNR down to 28 dB. From the User Profile Module, the system will know what type of device the request is originating from. In this case, we get the tuple h(28, 1, 20.6) from the table. Thus, we obtain the following. Then resulting µ = 11.60546 is exactly the expected compression ratio; this value is input to the ICM. Thus, after the compression phase the output frame ensures the image quality as well as the screen resolution This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Huu-Quoc Nguyen, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and quoc@khu.ac.kr; (2) Tien-Dung Nguyen, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and ntiendung@khu.ac.kr; (3) Van-Nam Pham, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and nampv@khu.ac.kr; (4) Xuan-Qui Pham, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and pxuanqui@khu.ac.kr; (5) Quang-Thai Ngo, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and nqthai@khu.ac.kr; (6) Eui-Nam Huh, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and johnhuh@khu.ac.kr. Abstract and 1 Introduction Abstract and 1 Introduction 2 Related Work 2 Related Work 3 Image Compression Module 3 Image Compression Module 4 System Architecture 4 System Architecture 5 Experimental Results 5 Experimental Results 6 Conclusions 6 Conclusions A Compression Results A Compression Results References References 4 System Architecture The proposed system provides an efficient interactive streaming desktop remote service for diversified client devices and dynamic network environments. The overall system structure is shown in Figure 4. Whenever a device requests a multimedia desktop streaming service, by the agent is installed on the client device, it helps transmit data included hardware and network environment parameters to User Profile Module (UPM). After that the UPM records the device codes and determines the required parameters, and then transmits them to the Dynamic Decision Estimated Module (DDEM). By analysis of historical data, the DDEM determines the most suitable compression ratio µ number for the device according to the parameters and the network condition. The µ number is transferred to the Image Compression Module (ICM) for use in the encode phase. Finally, the multimedia video file is transmitted to the mobile device through the service. 4.1 User Profile Module 4.1 User Profile Module The profile agent is used to receive the device hardware environment parameters and create a user profile. The device transmits its hardware specifications in XML-schema format to the profile agent in the cloud server, as shown in the Table 1. The XML-schema is meta-data, which is mainly semantic and assists in describing the data format of the file. The meta-data enables non-owner users to see information about the files, and its structure is extensible. However, any device that is using this cloud service for the first time will be unable to provide such a profile, so there shall be an additional profile examination to provide the test performance of the mobile device and sample relevant information. Through this function, the client device can generate an XML-schema profile and transmit it to the profile agent. The profile agent determines the required parameters for the XML-schema and creates a user profile, and then transmits the profile to the DDEM for identification. 4.2 Dynamic Decision Estimation Module with QoS Control 4.2 Dynamic Decision Estimation Module with QoS Control Figure 5 illustrates the DDEM used to estimate the value of σ. Specifically, the DDEM conducts this estimation after receiving the profile data from the User Profile Module. Based on multiple linear regression model in statistics, Based on multiple linear regression model in statistics, the DDEM determines the relationship between factors and predicts a suitable range for the number µ number. The problem is how to choose the best µ to obtain a suitable image size after the compression phase while also ensuring sufficiently high image quality and satisfying the image resolution demanded by the user. The ICM is used to obtain the compression data and calculate the corresponding P SNR values; these data are defined as historical data for the simulations, as shown in Table 3. In the current application, PSNRs of 25 to 50 dB are considered to be acceptable in terms of transmission quality. Therefore, before the regression analysis, we need to extract a list of values filtered by the condition P SNR ≥ 25. After we have a suitable range for the µ ratio, the system will analyze which µ number is most suitable for the compression phase. Specifically, let σ be the available bandwidth and θ the bandwidth threshold. If σ is greater than θ, the system can transmit the video content with the best possible quality without further processing. Otherwise, the next step is to determine appropriate values of P SNR and Compression Size to adapt to the current network conditions. In this case, the P SNR and Compression Size values are looked up in the estimation table; this table comprises a set of tuples of the form h(Resolution, P SNR, Compression Size), as described in the next section. Eventually, these tuples obtained is input to the estimate model, then the output µ value is asymptotic as well. The operation of QoS control is described algorithmically as Algorithm 3. 4.2.1 Proposed Estimation Model 4.2.1 Proposed Estimation Model In the present paper, we use multiple linear regression (MLR) to estimate the value of a variable based on the value of two or more other variables. The goal of MLR is to model the statistical relationship between response and predictor variables. A simple linear regression illustrates the relation between the dependent variable y and the independent variable x based on the regression equation According to the multiple linear regression model the dependent variable is related to two or more independent variables. The general model for k variables is of the form To simplify the computation, the multiple regression model in terms of the observations can be written using matrix notation. Using matrices allows for a more compact framework in terms of vectors representing the observations, levels of regression variables, regression coefficients and random errors. The model is in the form and when written in matrix notation we have where Y , X denote respectively the response values for all observations into a n-dimensional vector and predictor into a n×p+ 1 matrix. β denote the intercepts and slopes into a k + 1-dimensional vector and is an n × 1 vector of random errors. The first step in multiple linear regression analysis is to determine the vector of least squares estimators βb which gives the linear combination yb that minimizes the length of the error vector. Basically the estimator βb provides the least possible value to sum of the squares difference between yb and y. Now, since the objective of multiple regression is to minimize the sum of the squared errors, the regression coefficients that meet this condition are determined by solving the least squares normal equation In the present work, µ is considered to be the response variable and the Compression Size, Resolution and P SNR are used as predictor variables. In this case, we defined the Resolution as a value from 1 to 7 corresponding to one of seven common screen resolutions; these resolutions are listed in Table 3. Thus, from Eq. 3, we have the following linear model for this approach. In linear regression, one problem with least squares occurs when there are one or more large deviations, i.e. cases whose values differ substantially from the other observations. These points are called outliers. There are various methods for outlier detection [54]. As shown in Figure 6, using Cook’s distance which is commonly used of the influence of a data point when performing least squares regression analysis. Cook’s distance used to indicate data points that are particularly worth checking for validity. In this case, there are several outliers above the reference line; these are identified and removed from the model. After reducing the outliers, we obtain the following fitted regression model that represents the data adequately. The difference between the observed and predicted score, Y-Y’, is called a residual. We analyze a histogram that shows the range of the residuals and their frequencies, and a probability plot that shows how the distribution of the residuals compares to a normal distribution with matching variance. As shown in Figure 7, the new residuals plot looks fairly symmetric and the probability plot seems reasonably straight, suggesting that the data fits reasonably to normally distributed residuals. 4.2.2 Estimated Table 4.2.2 Estimated Table Specifically, the system will automatically adjust to increase or decrease the PSNR according to the available bandwidth σ representing the current network status. Then, by looking up the predictor variables P SNR, Resolution and Estimated Size in the estimated table, we can find the best tuple to input into the estimate model. Obviously, the µ value obtained here satisfies the client demands. For example, consider a case in which a thin client device sends a request to the system, and the available bandwidth σ is less than or equal to a predefined threshold value θ representing 80 `o the full bandwidth. The system will then decrease the PSNR value. The default is 30 dB. Assume we would decrease the PSNR down to 28 dB. From the User Profile Module, the system will know what type of device the request is originating from. In this case, we get the tuple h(28, 1, 20.6) from the table. Thus, we obtain the following. Then resulting µ = 11.60546 is exactly the expected compression ratio; this value is input to the ICM. Thus, after the compression phase the output frame ensures the image quality as well as the screen resolution This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv Authors: (1) Huu-Quoc Nguyen, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and quoc@khu.ac.kr; (2) Tien-Dung Nguyen, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and ntiendung@khu.ac.kr; (3) Van-Nam Pham, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and nampv@khu.ac.kr; (4) Xuan-Qui Pham, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and pxuanqui@khu.ac.kr; (5) Quang-Thai Ngo, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and nqthai@khu.ac.kr; (6) Eui-Nam Huh, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and johnhuh@khu.ac.kr. Authors: Authors: (1) Huu-Quoc Nguyen, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and quoc@khu.ac.kr; (2) Tien-Dung Nguyen, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and ntiendung@khu.ac.kr; (3) Van-Nam Pham, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and nampv@khu.ac.kr; (4) Xuan-Qui Pham, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and pxuanqui@khu.ac.kr; (5) Quang-Thai Ngo, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and nqthai@khu.ac.kr; (6) Eui-Nam Huh, Department of Computer Engineering, Kyung Hee University, 1 Seocheon, Giheung, Yongin, Gyeonggi, South Korea and johnhuh@khu.ac.kr.