AIRCC PUBLISHING CORPORATION
DYNAMIC SHAPING METHOD USING SDN AND NFV PARADIGMS
Department of Computer and Information Science, Seikei University, Japan
Traffic shaping controls communication traffic flow to prevent a specified communication rate from being exceeded. In conventional networks, the traffic shaping device is implemented at a predetermined location and only a communication flow passing through the device is targeted. If the traffic can be shaped dynamically on any selected communication flows at the optimal point only when necessary, it could use network bandwidths and packet relay processing capacity more efficiently and flexibly.
This paper proposes a dynamic shaping method using Software-Defined Networking (SDN) and Network Functions Virtualization (NFV) paradigms, which selects the optimal communication flows to be shaped, and the optimal shaping points dynamically. This paper also presented system configuration and functions for the proposed dynamic shaping, and the method to simplify the process of collecting the traffic data of each communication flow by SDN controller.
Traffic shaping, SDN, NFV.
Traffic shaping (hereafter “shaping”) is a form of bandwidth control. It controls communication traffic flow to prevent a specified communication rate from being exceeded. Any data that exceed the specified rate are stored in the communication device concerned and sent when the link concerned has spare capacity. It smooths a burst packet flow to produce a packet flow that is within the specified rate. The most common conventional way of shaping is to place-shaping devices in advance at predetermined points. These devices smooth out traffic according to a control policy specified for each application type or traffic type -. In most cases, traffic shaping is implemented in each network and only at points where there is regular traffic congestion, and only a communication flow passing through the device is targeted. Therefore, it has been difficult to apply shaping dynamically when and where it is necessary. If traffic can be shaped dynamically on any selected communication flows at optimal points only when necessary, it will be possible to use network bandwidths and packet relay processing capacity more efficiently.
This paper proposes a dynamic shaping method using Software Defined Networking (SDN)  and Network Functions Virtualization (NFV) - paradigms, which selects the optimal communication flows to be shaped and the optimal shaping points dynamically. This could make network resources much more economically available than conventional networks. The rest of this paper is organized as follows. Section 2 explains related works. Section 3 proposes a dynamic shaping method using SDN and NFV paradigms, which selects the optimal communication flows to be shaped and the optimal shaping points dynamically. Section 4 presents a system configuration for automating the proposed dynamic shaping. Section 5 confirms the operation of the proposed dynamic shaping with the evaluation system. Section 6 proposes the method to simplify the process of collecting the traffic data of each communication flow by the SDN controller, which is the key function for the proposed shaping method. Finally, Section 7 presents the conclusions. This paper is an extension of the study in Reference .
2. RELATED WORK
Reference  presents a comprehensive survey on the ON/OFF traffic shaping in the current Internet and summarizes the impacts of ON/OFF traffic on packet drop probability, real-time applications, and its interaction with TCP’s congestion control mechanism. Reference  overviews the features of Asynchronous Traffic Shaping (ATS) discussed in IEEE. Reference  proposes algorithms that allow flattening the utilization profile of network resources by optimizing network resources. Reference  proposes a QoE-aware traffic shaping method that is based on two QoE maximization metrics. A key benefit of this approach is that it calculates the optimal shaping rate to help clients to adjust its request for the subsequent segment quality level. Reference  compares the effect of traffic shaping and traffic policing on aggregate traffic dynamics especially stochastic properties on traffic time series. Reference  proposes a framework of traffic shaping in which the shaping filters are designed to interwork with statistical multiplexers that use FIFO buffers. Reference  explores a software-controlled architecture to implement a hierarchical control and management framework for QoS provisioning at network cores and proposes an optimization followed by a two-dimensional queue management policy, called Hierarchical Two Dimension Queuing (H2DQ).
In most of the above studies, traffic shaping is implemented in each network only at points where there is regular traffic congestion and the pre-deployed traffic shaper device, and only a communication flow passing through the device is targeted. Therefore, it has been difficult to apply shaping dynamically when and where it is necessary. For example, if a receiving link between network C and the terminal at the receiving side in Figure 1 is congested, the shaping has been implemented at the receiving side of Network C as Figure 1<1>. The bandwidth and the packet relay processing capacity of the networks A, B, and C related to the communication flow will be wasted.
3. PROPOSED DYNAMIC TRAFFIC SHAPING METHOD
3.1. Overview of Dynamic Traffic Shaping Method
The proposed dynamic shaping method is implemented with SDN- and NFV-based networks. It does not need to place-shaping functions at predetermined points in advance, as is the case conventionally. Instead, it detects the link congestion and dynamically selects communication flows to be shaped and the shaping points that are both optimal for resolving the congestion. It also places virtual shaping functions at the selected points and makes the selected communication flows pass through these points.
In the example of Figure 1, the shaping will be performed at the transmission side as in Figure 1 <2>. This can avoid wasteful use of network bandwidth and packet relay processing capacity, and consequently, can reduce the network cost.
Conventionally, it has been common to shape traffic not only for each link, but also for each application type and for each traffic type. If the shaping is to be applied to each application type in the conventional method, for example, the one that results in the greatest reduction in wasteful use of network bandwidth and packet relay processing capacity is selected. This paper discusses cases where traffic is shaped for each link but the same discussion applies to cases where traffic is shaped for each application type or for each traffic type.
If the dynamic traffic shaping is to be implemented in a conventional network, it would be necessary to introduce a system dedicated to measuring the traffic of each communication flow. If the traffic is shaped at optimal points, it would be necessary to place-shaping functions economically at all the points where communication flows pass through. However, the implementation of the above-mentioned functions is not economical. This paper proposes to take advantage of the features of the SDN and NFV, as shown in Figure 2. Specifically, SDN features make it possible as a basic function to measure the traffic data of each communication flow. Since the SDN controller keeps track of the route of each communication flow as a basic function, it is also possible to specify the route of each communication flow and perform shaping at an appropriate point. In addition, NFV features make it possible to place a shaping function of any capacity (even with a small capacity) at any point more economically than before. Additionally, by automating the dynamic shaping method, shaping can be performed quickly and maintenance operations can be significantly reduced.
3.2. Issues in Implementing The Proposed Dynamic Shaping Method
1) Trigger for shaping
It is assumed in this paper that a link is congested and shaping is executed if the link’s average usage rate exceeds Pmax (e.g., 0.7). This also could apply to cases where shaping is executed for each application type or for each traffic type, as mentioned in Section 3.1.
2) Selection of shaping points
Shaping at a point near the transmission side can reduce more network bandwidth and packet relay processing capacity than shaping at the congested point. Therefore, it is proposed to create a virtual shaping function with NFV features dynamically at a point near the transmission side when necessary, and the SDN controller changes the route so that the communications flow to be shaped will pass through that point.
3) Selection of the communication flow to be shaped
It is not efficient to shape all the communication flows that pass through the congested link. It is proposed to select the communication flow to be shaped as follows:
Among all the communication flows that pass through a link that is congested and requires traffic shaping, up to N (e.g., 10) fastest communication flows are selected as candidates for shaping.
As stated in 2), shaping at the transmission side can reduce the network bandwidth by L*V, where L is the link length between the transmission side and the congested link, and V is communication speed. As it is not easy to estimate the link length, it is proposed to use the number of hops (H), instead. The term x1 calculated by (1) is the reduced network bandwidth:
x1 = communication speed (V) × number of hops (H) (1)
Similarly, x2 calculated by (2) is the reduced number of packets to be relayed:
x2 = communication speed (V) ×number of hops (H) / packet length (P) (2)
Here, the packet length P is considered for the following reason. That is, even at the same communication speed, the shorter the packet length, the larger the number of packets to be relayed.
Finally, the communication flow with the largest x3 value calculated by (3) is selected to be shaped:
x3 = x1 * Cb + x2 * Cp (3)
where Cb and Cp are cost-coefficients which are used to calculate the cost of two different units at the same level.
For example, Flow b in Figure 3 is subject to shaping as it has the largest x3value.
4) Determination of shaping rate
It is desirable that the shaping rate that will resolve the congestion is selected. However, to guarantee a certain degree of quality of service, it is reasonable to limit the shaping rate so that the communication rate after shaping does not go below half of the original communication rate of each flow. If the congestion cannot be resolved by shaping the first flow with the largest x3 value, the flow with the next largest x3 value will be subject to shaping. This is continued until the congestion is resolved.
4. SYSTEM CONFIGURATION AND FUNCTIONS FOR THE PROPOSED DYNAMIC SHAPING
To execute the proposed dynamic shaping automatically, ‘the management orchestration’ is introduced in addition to the existing SDN controller and the equipment management system, which tries to minimize further additions to the existing system. The configuration of the system is illustrated in Figure 4. Terminals a, b and c communicate with the server individually. Their respective communication flows are called Flows a, b, and c. In this example, the link from the OpenFlow switch to the server is congested.
The management orchestration manages and executes the control scenario for dynamic shaping. Specifically, it observes the entire situation and executes the necessary processes. When it detects a congested link, the management orchestration instructs the SDN controller to collect data on the communication rate, the number of hops, and packet length for communication flows that pass through the congested link. The SDN controller gets these items of data from the OpenFlow Switch. However, if data on a large number of communication flows are to be collected, the performance of both the OpenFlow Switch and
the SDN controller degrades dramatically. To avoid this, the OpenFlow Switch monitors the traffic counter of each communication flow and notifies the SDN controller, in advance, of the communication flows whose traffic counters exceed a certain threshold. The SDN controller requests the OpenFlow Switch to send traffic data of only these communication flows.
Since the SDN controller knows the route in the network of each flow in advance, it sends the numbers of hops of these communication flows to the management orchestration. Based on the collected data, the management orchestration instructs the equipment management system to set the shaping control parameter and to provide information about the positions of shaping functions. It also instructs the SDN controller to change the route (route switching) so that the communication flows subject to shaping will pass through the specified shaping functions. The main functions that the management orchestration should have, as stated above, are summarized in Table 1.
The cooperation method from the selection of shaping function to route switching is illustrated in Figure 5, in which the number enclosed in squares indicates the process number.
5. CONFIRMATION OF THE OPERATION OF THE PROPOSED DYNAMIC SHAPING
5. 1 Design and Construction of an Evaluation System
Since the main aim of the evaluation is to confirm the operation of the management orchestration function, we implemented the SDN controller, the OpenFlow switch, and the equipment management system using substitute devices (Software router VyOS ). The evaluation tool (C#-based software program) was designed and constructed to implement the management orchestration functions proposed in Section 4. An example of the evaluation system is shown in Figure 6. It consisted of four VyOS nodes, three terminals, one control terminal, and one server. Terminals a, b, and c individually communicated with the server. It is assumed that the link from node 4 to the server will be congested. The shaping function was implemented using the VyOS function, not dedicated functions.
5.2 Results and Evaluation
The operation of the dynamic shaping was confirmed under the following conditions:
As the ratio of x3 value of flow a, x3 value of flow b, and x3 value of flow c is 6.5 : 11.8: 9.3, flow b with the largest x3 value among three flows is selected as the communication flow to be shaped in this example. In order to eliminate congestion, it is necessary to reduce the total communication speed from 17 Mbps to 14 Mbps (=20Mbps*Pmax), and therefore the shaping rate of flow b will be 4Mbps. Figure 7 shows changes in the measured speed from Node 4 to the server. As had been expected, the speed of Flow b was reduced to 4 Mbps.
6. METHOD TO SIMPLIFY THE PROCESS OF COLLECTING THE TRAFFIC DATA OF EACH COMMUNICATION FLOW BY SDN CONTROLLER
6.1 Basic Concept and Overview
The dynamic shaping method proposed in Section 4 requires the SDN controller to continuously collect traffic data at regular intervals for each communication flow, which is the key function for the proposed dynamic shaping method. As a result, the load on the SDN controller becomes very large and the performance of the SDN controller degrades dramatically. To avoid this performance degradation, the following policies can be considered:
<Policy 1>Reduce the number of communication flows for speed measurement.
For example, one method is to measure the speed continuously only for the communication flow with a higher speed measured after the start of communication. Another method is that OpenFlow Switch notifies the SDN controller of the communication flow in which its traffic counter value is greater than a certain value in advance, and the SDN controller queries the OpenFlow switch for only the traffic data of the notified communication flows.
<Policy >Speed measurement for communication flows after when the congestion is detected
Instead of constantly monitoring the speed of all communication flows, the speed of communication flows passing through the congestion link is measured after the congestion is detected.
<Policy 3>Increase the speed measurement interval for each communication flow.
<Policy 4>Stop the traffic data inquiry processing itself at the SDN controller.
The switch side periodically collects traffic data of communication flows to be monitored and estimates the communication speed (no inquiry from the SDN controller). Then, only when the speed has increased or decreased significantly compared to the previous cycle, the speed change is reported to the SDN controller. This is similar to the trap function of Simple Network Management Protocol (SNMP) , and the SDN controller instructs the switch in advance of the communication flow to be monitored.
In this paper, we propose a method based on the approach of policy 4, in order not to reduce measurement accuracy.
6.2 Extension of Openflow Specification
To implement the method proposed in Section 6.1, the following extensions to OpenFlow specification  are required:
(1) Add the following new fields to the Meter Modification message.
-Communication flow ID to be monitored
-Communication speed measurement cycle
-Degree of changes in communication speed and the packet length (Z0s, Z0p)
6.3 Message Sequence of Proposed Method
Figure 8 shows an example of the message sequence of the proposed method based on Sections 6.1 and 6.2. A broken line indicates a message which is used in the conventional method but becomes unnecessary this time, and a dashed line indicates a message which is added this time.
1) The SDN controller sends the switch a Meter Modification message containing the fields described in Section 6.1.
2) Unlike the conventional method, periodic exchange of the Flow_Stats multipart Request message, which is a request for statistical information for each communication flow, and the Reply message, which is a response to the request, is unnecessary. If the number of monitored flows is F for N times until the communication speed is notified, F*N messages processing could be eliminated in the entire SDN controller, and the processing load is greatly reduced.
This paper has proposed a dynamic shaping method using SDN and NFV paradigms. This method can select the optimal communication flows to be shaped and the optimal shaping points dynamically. This method can also avoid a many number of wasteful uses of network bandwidth and packet relay processing capacity as compared with the conventional method, and consequently, can significantly reduce the network cost. It has also presented a system configuration for automating the proposed dynamic shaping and the method to simplify the process of collecting the traffic data of each communication flow by SDN controller. The feasibility of the proposed method has been confirmed by an evaluation system.
It is required to evaluate the proposed dynamic shaping method experimentally in terms of resource utilization and stability comparing with the existing methods. It is also required to study how multiple SDN controllers and multiple equipment management systems collaborate.
CONFLICTS OF INTEREST
The author declares no conflict of interest.
We would like to thank Mr. Syunsuke SUZUKI for his help with the evaluation.
 Cisco, “Comparing Traffic Policing and Traffic Shaping for bandwidth Limiting”.
 T. Flach et al, “An Internet-Wide Analysis of Traffic Policing,” SIGCOMM’16, 2016.
 M. Marcon, M. Dischinger, K. P. Gummadi, and A. Vahdat, “The local and global effects of traffic shaping in the internet,” 2011 Third International Conference on Communication Systems and Networks (COMSNETS), pp.1-10, 2011.
 S. Bhaumik and S. Chakraborty, “Hierarchical Two Dimensional Queuing: A Scalable Approach for Traffic Shaping using Software Defined Networking,” NetSoft 2018.
 Y. Zhao, B. Zhang, C. Li, and C. Chen, “ON/OFF Traffic Shaping in the Internet: Motivation, Challenges, and Solutions,” IEEE Networks, Vol.31, Issue 2, pp.48-57, 2017.
 A. Grigorjew, F. Metzger, and T. Hobfeld, “Simulation of Asynchronous Traffic Shapers in Switched Ethernet Networks,” NetSys2019.
 A. M. Aladwani, A. Gawanmeh, and S. Nicolas, “Traffic Shaping and Delay Optimization in Demand Side Management,” 2013 UKsim 15th International Conference on Computer Modelling and Simulation.
 X.Liu and A.Men, “QoE-aware Traffic Shaping for HTTP Adaptive Steaming,” International Journal of Multimedia and Ubiquitous Engineering Vol.9,No.2,pp.33-44,2014.
 D. D. Simion and G. D. Horia, “ Traffic shaping and traffic policing impacts on aggregate traffic behavior in high speed networks,” 6th IEEE International Symposium on Applied Computational Intelligence and Informatics, 2011.
 A.Elwalid and D.Mitra, “Traffic shaping at a network node: theory, optimum design, admission control,” Proceedings of INFOCOM’97, April 1997.
 “OpenFlow Switch Specification Ver.1.5.1”, ONF (March 2015)
 D.Kreutz, F.M.V. Ramos, P. Verissimo, C. E. Rothenberg and S. Azodolmolky,“Software-Defined Networking: A Comprehensive Survey,” Proceedings of the IEEE, Vo.103, Issue 1, pp.14-76, Jan.2015.
 “Network Functions Virtualization – An Introduction, Benefits, Enablers, Challenges and Call for Action,” ETSI Introductory White Paper. https://portal.etsi.org/NFV/NFV_White_Paper.pdf
 “Network Functions Virtualisation (NFV); Architectual Framework,” ETSI GS NFV 002 v1.2.1, Dec. 2014.
 R.Mijumbi, J.Serrat, J.Gorricho, N.Bouten, F.D.Trurck and R.Boutaba, “Network Function Virtualization: State-of-the-art and Research Challenges,” IEEE Communications Surveys & Tutorials, Vol. 18, Issue 1, pp.236-262, 2016.
 M.Bouet, J.Leguay, and V.Conan, “Cost-based placement of virtualized Deep Packet Inspection functions in SDN,” 2013 IEEE Military Communications Conference, pp.992-997, Nov. 2013.
 VyOS, https://www.vyos.io/2020.08.13
 IETF RFC3416 “Version 2 of the Protocol Operations for the Simple Network Management Protocol (SNMP)”https://tools.ietf.org/html/rfc3416
 S.Suzuki and S.Kuribayashi, “SDN-based Dynamic Traffic Shaping Method ,” Proceeding of INNOV2020, Oct. 2020.
Shin-ichi Kuribayashi received the B.E., M.E., and D.E. degrees from Tohoku University, Japan, in 1978, 1980, and 1988 respectively. He joined NTT Electrical Communications Labs in 1980. He has been engaged in the design and development of DDX and ISDN packet switching, ATM, PHS, and IMT 2000, and IP-VPN systems. He researched distributed communication systems at Stanford University from December 1988 through December 1989. He participated in international standardization on ATM signaling and IMT2000 signaling protocols at ITU-T SG11 from 1990 through 2000. Since April 2004, he has been a Professor in the Department of Computer and Information Science, Faculty of Science and Technology, Seikei University. His research interests include resource management for NFV and SDN-based networks, QoS control, traffic control for cloud computing environments, IoT traffic management and, green network. He is a member of IEEE and IEICE.