Time Awareness in Software Defined Networking

2015 European Conference on Networks and Communications (EuCNC)

Time Awareness in Software Defined Networking Padraig O Flaithearta Discipline of IT, College of Engineering & Informatics, NUI Galway, Galway, Ireland [email protected]

Hugh Melvin Discipline of IT, College of Engineering & Informatics, NUI Galway, Galway, Ireland [email protected]

implement multiple protocol standards/algorithms, instead they would just accept instructions from SDN controllers. SDN thus enables more dynamic and flexible network provisioning and will assist applications to achieve the required QoS. Within this context, we argue that precise and verifiable timing, and more generally, time awareness can greatly assist in this process. Information and Communications Technology (ICT) systems however do not readily support this time awareness functionality. Instead, much of the developments in recent decades have focused on delivering better throughput and fairness, both on end devices and networks, at the expense of temporal determinism. Our proposal lies within the broader question being addressed by the Time Aware Applications, Computer, and Communications Systems (TAACCS) interest group set up in 2014 [2]. TAACCS aims to highlight time awareness opportunities, identify a range of challenges to time awareness, and pursue a cross disciplinary approach to meet the challenges. According to Cisco [3], there will be 50 billion devices connected to the Internet by 2020. Currently, most applications are human centric and based on smartphones/tablets as hubs for connectivity, though an increasing number of smart devices and sensors are connecting to the Internet. This latter trend is resulting in more complex systems and interactions [4] and is expected that the IoT will create $14.4 trillion in economic value by 2022, while several industry leaders, including Amazon, Cisco, GE, and Qualcomm, all believe that the Internet of Everything (IoE) could have a bigger impact on the world than the Internet that preceded it [5]. Such growth as well as the expected growth in more conventional streaming will result in QoS challenges for Real Time Communications (RTC) as the network is required to deliver a range of traffic types. In this context we believe that SDN with time awareness can help address some of these. The remainder of the paper is structured as follows; Section 2 provides some further background information on Software Defined

Abstract— With the advent of the Internet of Things (IoT), networks will need to handle so-called big data from a range of new sources including Smart Transportation, Smart Cities, Industrial Internet, and Cyber Physical Systems. Increasingly connected over wireless, these devices will produce and consume data for Machine to Machine (M2M) communication and will share networks with more conventional user-generated data from smartphones and tablets. All of this new traffic will emerge along with users increasingly high expectations for Quality of Service (QoS) in traditional Real Time Communications (RTC). Headed by the Open Networking Foundation (ONF), Software Defined Networking (SDN) is an emerging network architecture that will hopefully facilitate such high-bandwidth, dynamic applications with differing QoS requirements. The core concept of SDN is that network control is abstracted away from individual devices towards centralized computing devices. In this paradigm, network intelligence is centralized in software-based SDN controllers, in effect, rendering the network as a single logical switch. In this paper, we explore how introducing time awareness in an SDN network can help further optimize QoS. Keywords—SDN, NTP, TAACCS, E-Model, QoS, VoIP I.

Peter Pocta Department of Telecommunications and Multimedia, FEE, University of Zilina, Zilina, Slovakia [email protected]

INTRODUCTION

Software Defined Networking (SDN) is an enabling technology that will help meet growing demands in terms of traffic throughput as well as differentiated QoS. It realizes a network architecture whereby network control is decoupled from forwarding and is ‘directly programmable’ which enables underlying infrastructure to be abstracted for applications and network services, which can then ‘treat the network as a logical or virtual entity’ [1]. SDN also simplifies networking devices, as these no longer need to understand or

580

2015 European Conference on Networks and Communications (EuCNC)

Networking and how time awareness can improve SDN design/implementation. Section 3 provides details of a case study at the PEL research group at NUI Galway in which significant QoS improvements are shown in a VoIP-over-WiFi scenario using a time awareness concept in an approach that reflects an SDN architecture. Section 4 concludes the paper. II.

effectively use the above information and help ensure that applications that require certain QoS treatment are prioritized accordingly. For example, a cross layer approach can translate dynamic delay information and QoS requirements into dynamic MAC priorities that are communicated back down to the network devices.

BACKGROUND 3.

Timing Distribution: In the absence of accurate time sources such as GPS, timing must be delivered across networks. This distribution is typically done using protocols such as NTP/PTP and is impacted greatly by asymmetric delays. An SDN with Time Awareness can help to ensure that timing traffic is treated appropriately, thus minimizing any asymmetries and improving the level of synchronization. As such, the SDN manages RTC traffic flows appropriately, whereby timing traffic may constitute traffic that also needs very high priority treatment. One key challenge in the context of point 3 above is that distribution of timing across the last hop where wireless networks such as WiFi are used is quite difficult due to asymmetries. Related and patented work at NUI Galway [6], provides a mechanism that implements and achieves time synchronization accuracy using NTP over WiFi 802.11 networks similar to that achievable over wired networks. This significant improvement dynamically determines the delays incurred by time messages as they traverse a wireless link. These delays may then be used to reduce errors associated with non-deterministic time message latencies, which leads to greatly improved synchronization accuracy. In the following section we describe a case study whereby NTP is used to provide sub-millisecond time synchronization accuracy on all nodes participating in multiple RTC sessions through a WiFi access point. Real-time delays are determined and used to dynamically alter wireless MAC delays via a cross layer approach to optimize QoS and QoE perceived by the end user.

Although SDN was originally designed to improve wired backbone networks by improving the performance of switches, we expect that its benefits will also be required over last hop wireless networks. This is because most of the growth in connectivity listed above will be over wireless networks. Wireless, and especially mobile networks are often characterized by very dynamic network conditions which tend to yield decreased QoS and Quality of Experience (QoE) when compared to well provisioned and largely symmetric wired connections. In the context of mobile networks topology and load, information from multiple network domains (e.g., mobile and Wi-Fi access networks), combined with application requirements and user context, can enhance mechanisms such as mobile data offloading and vertical (cross-technology) handovers. Especially in the case of Wi-Fi, this can enable efficient management of wireless resources (e.g., reducing wireless interference through appropriate channel selection), Wi-Fi access sharing, and content caching, which can further improve the offered QoS/QoE and service performance. We argue that the expected massive increase in M2M traffic over computer networks along with the expected growth in HD video streaming will cause further challenges that can benefit from SDN approaches and that in turn SDN approaches will benefit greatly from time awareness. Timing in the broadest sense refers to time, frequency and phase which though distinct, are related. In particular SDN can benefit from time awareness in the following ways: 1. Time Synchronization: By providing precision time e.g. based on Universal Coordinated Time (UTC), on all nodes in a network to millisecond level (or better if needed), this can yield precise real-time delay information. Such dynamic information, if fed in real-time to SDN controllers, can help to meet QoS requirements. Determining and feeding such delay information back to the controller is however a non-trivial challenge, that needs careful design and support. 2.

III.

CASE STUDY

Although the IEEE 802.11 (WiFi) standard has a QoS support extension described in IEEE 802.11e that offers prioritization between various traffic access categories, the case study builds on IEEE 802.11e by providing for traffic prioritization within access categories. A dual approach of simulation and real-world experimentation is taken. The former is used to validate the core idea whereas the latter is used principally to assess the technical feasibility of the approach. Regarding the latter, as part of an SDN based proof-of-concept (PoC) test-bed (Figure

Time Awareness on SDN controllers: Time awareness on controllers can

581

2015 European Conference on Networks and Communications (EuCNC)

1), an intelligent Access Point (iAP – Figure 2) is presented which integrates a number of key features that operate in real-time.

each session. The R-factor value derived for each session is then passed to the Prioritization module which runs a prioritization algorithm to decide which, if any VoIP sessions should be prioritized.

Fig. 1 Proof-of-concept test-bed Fig. 2: iAP Architecture

The iAP consists of a number of modules that combine to assess in real-time the QoS of all VoIP sessions in a network and then use this information to optimize QoS/QoE. The calculation of accurate one-way delays requires the clocks of all end nodes in the test-bed being synchronized with millisecond accuracy. As outlined in section 2, the challenges of implementing time synchronization over WiFi wireless networks is an active research topic at the PEL research group at NUI, Galway, and details of some of this work can be found in [6]. The iAP (illustrated Figure 2) dynamically and in real-time performs the following at fixed intervals: 1. Identify individual VoIP sessions 2. Calculate one-way delay for each session, including wired/wireless components of the delay, termed intra-one-way delay 3. Calculate each-way R-values (quality perceived by the end user estimated by the E-Model ‎[7]) for each session using one-way delays from step 2. 4. Run a prioritization algorithm for VoIP sessions (based on R-values) 5. Implement prioritization on AP downlink amongst sessions via modification of the IP DSCP field which is interpreted appropriately at the MAC layer. Note that the mechanism operates on the downlink only as much research has highlighted the so-called AP bottleneck as the main source of delay/loss. These actions can be controlled from a remote management device as shown in Figure 1 where a remote station controls the iAP via SSH. As data packets enter the AP, a packet capture application scans for RTCP SR/RR packets. When RTCP packets are found, the AP timestamp is noted and they are passed up to a Delay Calculation module which identifies unique VoIP sessions and calculates the intra-one-way, and one-way delays for each session. The delay is calculated for both directions, namely uplink and downlink separately. These delay values are then passed to an E-Model based QoE module to generate an R-factor score for

If a VoIP session is to be prioritized, the assigned ID of the session is passed to the Traffic Reclassification module which modifies (or mangles) the DSCP value in the IP header of VoIP data packets. This ensures that when these packets arrive at the MAC layer they are redirected appropriately to prioritize the session in question. In order to implement prioritization for concurrent VoIP sessions, a three-tiered categorization mechanism (Figure 3) is used in the PoC whereby all sessions initially reside in the lowest category, and thereafter, certain sessions may be promoted (prioritized) to a higher priority category or demoted (de-prioritized) where possible, based on their R-factor score. A background traffic category also exists in order to cater for non-VoIP traffic. The three Multimedia Categories are termed MC1, MC2, and MC3, where MC1 has the highest priority and MC3 has the lowest priority, along with a background access category (BK). All VoIP traffic resides within the three MCs. These four categories were chosen so that existing IEEE 802.11e equipment could be used in the PoC as IEEE 802.11e implements a four category system.

Fig. 3 Multimedia Categorization System

582

2015 European Conference on Networks and Communications (EuCNC)

show how the approach is aligned with SDN techniques for more centralized control. Regarding ongoing and future work, the authors are participating in the recently formed EU COST Action ACROSS [9]. The action is in part looking at issues of reliability and QoS in suture Internet. The authors with other are building on the research presented in this paper and aim to apply the concept of time awareness in SDN across large scale networks and to examine its potential for RTC – especially VoIP and gaming.

When a VoIP session is chosen for prioritization or de-prioritization, the iAP needs to implement this. The design approach utilizes the Netfilter framework to modify the Diffserv Code Point (DSCP) value in the IP header of all downlink packets (RTP) belonging to the session. A mapping exists between DSCP values and the IEEE 802.11e MAC traffic categories, and this process is utilized in the design to implement re-classification between MC1 – MC3. This has the effect of ensuring that all voice packets for a session will follow the categorization framework described in the previous section. Further details can be found in [8]. Results show that the approach is both valid and feasible. The approach is also very much aligned with the emerging SDN paradigm described earlier. IN the above example the iAP acts as a centralized controller that monitors all RTC traffic, and can dynamically optimize QoS for each RTC session in real-time. However, we envisage that it can be extended to more complex mesh WiFi networks and is scalable. As shown in Figure 1, a remote management station can also be used to control the iAP. Whilst Figure 2 realizes the intelligence within the AP, we also experimented with offloading the intelligence to the remote management station. In this approach, the delay calculation, QoS estimator and prioritization modules are remote and we simply pass down the prioritization decisions for each session ID to the AP which becomes a slave AP as illustrated in Figure 1. Such a system could then be extended whereby the management station becomes an SDN controller for a larger scale mesh network. In this context, we have also examined scalability issues on AP hardware. In particular we have examined the extent of mangling that can be significant for APs that manage multiple concurrent RTC flows. Analysis shows that the processing latencies are negligible using conventional hardware, relative to the Mouth to Ear (M2E) latencies being considered. Furthermore we examined the scalability of the delay calculation mechanism using RTCP. Essentially, the sniffing and analysis of RTCP traffic is undertaken first/last hop of our small scale network (a single WiFi AP). However the mechanism described may operate on a full mesh network as long as the following two conditions are met; 1 – all session endpoints and intermediate iAPs are synchronized, and 2 – the session traffic is transported using RTP along with RTCP control. IV.

ACKNOWLEDGEMENT This work has been supported/partially supported by the ICT COST Action IC1304 - Autonomous Control for a Reliable Internet of Services (ACROSS), November14, 2013 – November 13, 2017, funded by European Union.

REFERENCES [1]

[2]

[3]

[4] [5] [6]

[7]

[8]

[9]

CONCLUSION

This paper demonstrates the feasibility of time awareness through development of a proof-ofconcept implementation, based on the WiFi QoS protocol 802.11e. More generally, we examine the extent to which time awareness across the full ICT infrastructure can assist in better QoS and QoE and

583

Software Defined Networking: The New Norm For Networks, ONF White Paper, Open Networking Foundation, April 2012 M. Weiss, J, Edison, C. Berry, L. Goldin, B. Iannucci, E. Lee, K. Stanton, “The Case for Cross Disciplinary Research on Time Aware Applications, Computers and Communication Systems (TAACCS)”, White Paper, http://tf.nist.gov, September, 2013 J. Bradley, J. Barbier, D. Handler, “Embracing the Internet of Everything to Capture Your Share of $14.4 Trillion”, White Paper, Cisco, 2013 M. Royer, “The Internet of Things (IoT) A Trends White Paper”, Bellvue College, 2013 “Everything Changes with the Internet Of Everything”, Forbes, August 2013 Shannon, H. Melvin, R. O hOgartaigh, A. Ruzzelli, “Synchronization Challenges with Future Smart Grid Networks”, Intl. Conf. on Smart Grids, Green Communications and ITaware Technologies, ENERGY 2011, Venice, Italy, May 2011 ITU-T Recommendation G.107, “The EModel, a computational model for use in transmission planning”, December 2011 P. O Flaithearta, H. Melvin, M. Schukat, “A QoS Enabled Multimedia Access Point”, ACM International Journal of Network Management (IJNM), In Press, 2015 EU COST IC 1304- Autonomous Control for a Reliable Internet of Services (ACROSS) http://www.cost.across.nl/