Priority-based Scheduling Policy for OpenFlow Control Plane

The Board of Examiners has unanimously approved this dissertation, submitted by Mr. Piyawad Kasabai, in partial fulfillment of the requirements for the Doctor of Philosophy in Computer Science at Mahasarakham University. Mahasarakham University has granted approval to accept this dissertation in partial fulfillment of the requirements for the Doctor of Philosophy in Computer Science. The purpose of PSO is to give higher priority to OpenFlow control messages in the in-band control network.

Additionally, PSO provides different priorities for OpenFlow control messages based on content/services (traffic types) and/or clients for both in-band and out-of-band control network.

Figure 25 OpenFlow packet of a high priority data traffic on a control link (CL-1): a) Packet loss, b) Delay ...................................................................................................

INTRODUCTION

For communication between control and data planes, out-of-band control network will be evaluated in this work. However, this work takes the DSCP into the design mainly to mitigate the delay of higher priority traffic only, which does not cover all QoS parameters. Data plane A plane related to data traffic that is detected in hardware and is responsible for forwarding packets.

Data traffic Traffic related to data packets, such as ARP packets, TCP packets, UDP packets, and so on.

BACKGROUND AND RELATED WORK

In general, both centralized and distributed controllers can cause delay issues, namely inbound and outbound delays. The inbound delay (or inbound latency) occurs when a switch generates packet-in messages and sends them to a controller. 11] have found that both delays can lead to inefficiency of the connection between the switch and the controller. The flow table is built into the switch hardware using Ternary Content Addressable Memory (TCAM).

The cookie field contains the cookie of the flow entry, which caused the OpenFlow message to be sent to the controller. The buffer_id and path fields are given the same in the package-in message. The OFPT_FLOW_MOD message (as shown in Figure 7) is used to modify flow entry in the flow tables.

The flow tables of the OpenFlow switch are used to serve the incoming data packets. Then the switch sends the packet-in message to the controller to request an action or a new flow entry that is stored in the flow tables. A package can be associated with a flow entry in the flow tables using one or more package header fields.

They manage the power control of various switches/routers in the network from the centralized controllers. Database management involves managing all the tables in the network devices with data synchronization. In the Internet Protocol (IP), Differential Services (DiffServ) [42] is a traffic control architecture, which relies on the 8-bit DS field (instead of the legacy Type of Service (ToS) field [43]) in the IP -network. header.

In [10], the authors proposed queuing and fault recovery features for OpenFlow in an in-band control network.

RESEARCH METHODOLOGY

The goal of the ns-3 project is to develop a preferred open simulation environment for network research. Delay is a crucial index of the operational efficiency of SDN networks, especially for real-time applications (such as Voice over IP). In case of table miss in OpenFlow switches, a data packet will be encapsulated in an OpenFlow packet-in message.

During network congestion, this OpenFlow packet-in message can cause an increase in delay. D1 is the hop delay, which counts the time from when switch S1 sends a packet-in message to the controller until switch S1 receives a packet-out message (i.e., acquiring forwarding rules). This time includes the processing time at both switch S1 and the controller, queuing delay at both switch S1 and the controller, and the transmission time of the packet-in and packet-out messages.

The controller can look at the network end-to-end while making instructions to the switches because it has a full physical and logical view of the network topology. Packet loss is defined as the fraction of total transmitted packets that have not been received at the receiver. According to TCAM operations, the OpenFlow module in ns-3 [45] considers the concept of virtual TCAM to estimate the average search time of flow tables.

This topology is suitable for evaluating the situation in networks that are unsuitable, such as the competition between control traffic in an out-of-band control network (detailed in Chapter 5). In this work, each simulation will be run 50 times using a different Random Number Generator (RNG) seed to obtain the average results, quoted with error bars with respect to 95% confidence intervals.

Table 6 Metrics/Parameters of IETF IPPM RFCs Category IETF IPPM RFCs [50]

DESIGN

Thus, the unmatched packets are forwarded to the next hop SDN interface until they reach the controller. For the out-of-band control network, the SDN interface (or an SDN port) of an OpenFlow switch is a specific port of that switch, which connects directly to the OpenFlow controller [55]. For the in-band control network, some dedicated ports are deployed on each switch to pass OpenFlow control messages to the controller.

Otherwise, some SDN interfaces in the in-band control network may indirectly connect to the controller through other switches. As mentioned earlier, these SDN interfaces in an in-band network could suffer from latency due to competition between OpenFlow control messages and data packets carried over the same interface. In an in-band control network, data packets can share the same connection with OpenFlow control messages.

Finally, Rule #4 gives the lowest priority to data packets and sets their DSCP headers to be equal to the DSCP value within the data packets. These DSCP values of OpenFlow control messages and data packets will then be considered by a packet scheduler to schedule traffic according to their priorities (such as using WFQ). The last rule (Rule #5) gives the lowest priority to data packets and sets their DSCP headers to be equal to the DSCP value within the data packets.

After passing the traffic classifier, packets are differentiated according to the rules in PMT. In the first case, the packet scheduler schedules the packets according to the DSCP values and scheduling mechanisms (defined by the network administrator).

Figure 17 SDN-enabled Ethernet switch Source: [55]

PERFORMANCE EVALUATION

Experiment 1: a PMT for an out-of-band control network to give a priority to a specific service. This is to allow all configuration commands from network administrators and OpenFlow packet-in/packet-out messages from real-time services (UDP port 20000) to have higher priority than other traffic. Experiment 2: a PMT for an out-of-band control network to provide a priority for a specific user/customer.

Experiment-1: a PMT for an out-of-band control network to prioritize a specific service. In traditional OpenFlow, with a high load (NL > 0.8), some buffered high-priority packets are then dropped after their expiration, because the OpenFlow control messages on CL-1 are dropped. So, our PSO can help the high-priority data flow to obtain the lowest-delay forwarding rules under network congestion on the control link.

Experiment-2: A PMT for an out-of-band control network to prioritize a specific user/customer. However, at high load (NL > 0.8), the congestion causes a significantly high packet loss rate in traditional OpenFlow. Thus, our design can help the high-priority data stream to acquire forwarding rules with less delay under network congestion on the control link.

In the case of control link congestion, traditional OpenFlow would cause a serious problem for a given privileged user/client. For the in-band control network, we designed PSO to prioritize OpenFlow messages over data traffic.

Figure 23 OpenFlow packet of a high priority data traffic on a control link ( CL - 1 ): a )

CONCLUSIONS AND FUTURE WORK

However, the design of some in-band control network mechanisms are still an open research issue. This work has evaluated the OpenFlow protocol and proposed a new design of the OpenFlow protocol successfully. This work has established a set of key performance evaluation criteria and performance metrics/parameters to evaluate the proposed solution.

When the control connections are under high load, traditional OpenFlow causes latency issues, slowing down sensitive services and privileging the user/client. Although several successes have been claimed in this thesis, there are also some weaknesses. The following aspects discuss some limitations of this thesis and the issues that would be explored in future work.

However, these simple scenarios are useful to evaluate the situation in network congestion in SDN. However, due to the limitation of ns-3 modules and the unfinished design of inner tire mechanisms, this thesis has not yet experimented to evaluate in the inner tire control environment. So, one of the future directions could be to propose and experiment on the mechanisms for the in-band networks on the open research issues.

However, after simulation, prototyping and testing on the real test bed would be a good idea. So, the next step could be to implement and evaluate the PSO on the real testbed using a tool, such as the GENI testbed [59].

Aditya, "Unraveling the Complexity of Network Management," v USENIX Symposium on Networked Systems Design and Implementation, str. Figueiredo, et al., "Implementing Quality of Service for the Software Defined Networking Enabled Future Internet," v Third European Workshop on Software Definirana omrežja, Budimpešta, 2014, str. 34; Latency in Software Defined Networks: Measurements and Mitigation Techniques," ACM SIGMETRICS Performance Evaluation Review, vol.

Bao, “A Survey of Software-Defined Networking and OpenFlow: From Concept to Implementation,” IEEE Communications Surveys Tutorials, vol. 34;Design and Implementation of a Routing Control Platform", in Conference on Networked Systems Design & Implementation, 2005, pp. Wetherall, "Towards an active network architecture", in DARPA Active Networks Conference and Exposition, New York, NY, USA, 2002, p.

Erickson, "The Beacon Openflow Controller," in ACM SIGCOMM Workshop on Hot Topics in Software Defined Networking, New York, NY, USA, 2013, pp. Yedavalli, "Network virtualization and resource allocation in OpenFlow-based wide area networks," in IEEE International Conference on Communications (ICC), Ottawa, 2012, pp. Rexford, "A NICE way to test OpenFlow applications," in USENIX Symposium on Networked Systems Design and Implementation (NSDI 12), San Jose, CA, 2012, pp.

Okabe, “A packet-in-message filtering mechanism for control plane protection in open-flow networks,” in ACM/IEEE Symposium on Architectures for Networking and Communications Systems (ANCS), Los Angeles, California, USA, 2014, pp. Schmid, "A self-organizing distributed and in-band SDN control plane," in International Conference on Distributed Computing Systems (ICDCS), 2017, pp.

BIOGRAPHY