BeaQoS: Load balancing and deadline management of queues in an OpenFlow SDN switch
Introduction
Software Defined Networking (SDN) is revolutionizing the networking industry by enabling programmability, easier management and faster innovation [1], [2]. These benefits are made possible by its centralized control plane architecture which allows the network to be programmed and controlled by one central entity.
The SDN architecture is composed both of SDN enabled devices (switches/routers)1 and of a central controller (SDN controller). An SDN device processes and delivers packets according to the rules stored in its flow table (forwarding state), whereas the SDN controller configures the forwarding state of each SDN device by using a standard protocol called OpenFlow (OF) [2]. The SDN controller is responsible also to build the virtual topology representing the physical topology. The virtual topology is used by the application modules that run on top of the SDN controller to implement different control logics and network functions (e.g. routing, traffic engineering, firewall actions).
Currently the Quality of Service (QoS) management in OF is quite limited: in each OF switch one or more queues can be configured for each outgoing interface and used to map flow entries on them. Flow entries mapped to a specific queue will be treated according to the queue’s configuration in terms of service rate, but the queue’s configuration takes place outside the OF protocol. For example, the queue’s service rate cannot be modified by OF.
Supposing that a flow is traversing a chain of queues from the source to the destination node, and that the flow data rate increases, a possible consequence is that queues increase their occupancy, and a bottleneck may be generated with consequent network congestion. The impossibility to change the bottleneck queue’s service rate through real-time OF directives can lead to a severe performance degradation for the flows traversing that queue because, without a proper rate assignment, it is very difficult to guarantee Quality of Service requirements to the flows [3].
A possible solution to mitigate the performance degradation involves the re-routing of the flows experiencing a violation of deadline constraints (e.g. the flows that are totally received beyond the fixed time constraint) [4] on less congested paths or queues. The underlying idea is that, since we cannot change the service rate of the queues, we act on the ingress traffic, moving a subset of flows on different paths or queues in case of need. In order to be 100% compatible with current OF hardware, we impose no changes to OF specifications and directives. Instead we propose to modify one popular SDN controller: Beacon [5]. The proposed solution, BeaQoS, applied to a single SDN switch, is an extension of our previous work presented in [6]. Our new updated controller will receive statistics about queues, flows and ports from OF switches and will compute an estimation of the flow rates and of the packet loss of the queues. Based on customizable policies, BeaQoS will be able to select a subset of flows experiencing congestion over the bottleneck queue and to re-route them on another and less congested queue, so improving the switch performances. The action of flow re-routing may be exploited not only for deadline management but also for efficient queue load balancing. On the other hand load balancing is often seen as an action to prevent congestion and, consequentially, to limit and delay performance detriment.
The remainder of this paper is structured as follows. We describe related works on this field in Section 2. Concerning the main contributions of the paper:
- –
We explain the motivations that lead to consider multi-queue interfaces with variable service rate to support deadline management in Section 3;
- –
We describe the basic idea concerning the re-routing mechanisms introduced in this paper in Section 4.1, where we also show how it can be usefully applied in case of multi-core architectures and load balancing issues among queues;
- –
We describe the modifications of the Beacon controller required to implement re-routing in Section 4.2;
- –
We propose five effective re-routing strategies in BeaQoS: two of them aimed at improving deadline management and three of them aimed at balancing the load among queues in a SDN switch in Section 5.
We show the performance analysis of our proposed algorithms in Section 5. We report a discussion about the obtained results together with the conclusions in Section 7.
Section snippets
Related works
Despite traffic engineering (TE) approaches are often ruled by MPLS-TE [7], [8], the ability of the SDN controller to receive (soft) real-time information from SDN devices and to make decisions based on a global view of the network, coupled with the ability of “custom”-grained flow aggregation inside SDN devices, makes TE one of the most interesting use cases for SDN networks.
Global load balancing algorithms are proposed in [9] that addresses load-balancing as an integral component of large
Motivations
Some approaches consider a single queue for each outgoing interface. In order to support QoS mechanisms and traffic differentiation, it is common to configure multiple queues in advance [3]. The importance of traffic differentiation is highlighted by the first group of simulations (Table 1) reported in the following.
Flow entries mapped to a specific queue will be treated according to that queue’s configuration in terms of service rate. Most of the previously mentioned approaches assumes the
General idea
Although the design and implementation of a new OpenFlow directive able to configure the queues’ service rate would be the best solution in terms of performances, this choice would come up with a main drawback: it would be totally incompatible with current OF switches that would not take any benefit from the directive.
For this reason we propose an alternative solution totally compatible with current OF switches. The underlying idea is shown in Figs. 1 and 2.
Let us suppose that, during the
Re-routing strategies analysis
In this section, we present two main scenarios in which we compare different proposed re-routing algorithms to find the most efficient solution. The first scenario deals with the problem of the priority flows that must be served within a specific deadline, as introduced in Section 3. The second one faces the issue of balancing the load among different queues in a single SDN node.
Scaling performances
Concerning statistics (see Table 4) acquisition: the types of messages sent by the controller are flow, queue and port requests that are used to gather information about port rates, queue rates and individual flow statistics. The controller receives three statistic replies, one for ports, one for queues and one dedicated to all flows traversing the OpenFlow switch in a given instant.
Since the maximum information sent through the Ethernet frame is 1500 byte, each flow statistics reply message
Conclusions
The impossibility to configure the service rate of the queues in a OpenFlow switch through an OF directive is a limitation that could reduce the quality management capabilities in an SDN network but it is a fact for now.
In this paper, exploiting the re-routing mechanism, we propose a method able to provide a basic deadline management support and an efficient queue balancing without any modification of OpenFlow specifications and switches. We present BeaQoS, an updated version of the Beacon
Luca Boero was born in Genoa, Italy in 1989. In 2012 he got his Bachelor Degree in Telecommunication Engineering at the University of Genoa; in 2015 he achieved a Master Degree in Multimedia Signal Processing and Telecommunication Networks with a thesis on Quality of Service in Software Defined Networking. From November 2015 he is a Ph.D student at the University of Genoa. His main research activities concern networking and SDN.
References (28)
- et al.
Ethane: Taking control of the enterprise
Proceedings of the 2007 Conference on Applications, Technologies, Architectures, and Protocols for Computer Communications
(2007) - et al.
Openflow: enabling innovation in campus networks
SIGCOMM Comput. Commun. Rev.
(2008) QoS Over Heterogeneous Networks
(2007)- et al.
Better never than late: meeting deadlines in datacenter networks
Proceedings of the ACM SIGCOMM 2011 Conference
(2011) The Beacon Openflow Controller
Proceedings of the Second ACM SIGCOMM Workshop on Hot Topics in Software Defined Networking
(2013)- et al.
Management of non-conformant traffic in openflow environments
Performance Evaluation of Computer and Telecommunication Systems (SPECTS), 2015 International Symposium on
(2015) - D. Awduche, L. Berger, D. Gan, T. Li, V. Srinivasan, S. G, Rsvp-te: extensions to rsvp for lsp tunnels, 2001, (RFC...
- et al.
Load optimal mpls routing with n + m labels
INFOCOM 2003. Twenty-Second Annual Joint Conference of the IEEE Computer and Communications. IEEE Societies
(2003) - et al.
Aster*x: load-balancing as a network primitive
Plenary Demo, 9th GENI Engineering Conference
(2010) - et al.
Openflow-based server load balancing gone wild
Proceedings of the 11th USENIX Conference on Hot Topics in Management of Internet, Cloud, and Enterprise Networks and Services
(2011)
Achieving high utilization with software-driven wan
Proceedings of the ACM SIGCOMM 2013 Conference on SIGCOMM
Traffic engineering in software defined networks
INFOCOM, 2013 Proceedings IEEE
Splendid isolation: a slice abstraction for software-defined networks
Proceedings of the First Workshop on Hot Topics in Software Defined Networks
Cited by (41)
Traffic aware dynamic load distribution in the Data Plane of SDN using Genetic Algorithm: A case study on NSF network
2022, Pervasive and Mobile ComputingCitation Excerpt :The multi-controller deployment algorithm proposed in Jiugen et al. [21], Ma et al. [22], Wang et al. [14] balances the load among the controllers placed in different regions but not among the switches. Re-routing the traffic is a solution, where Hu et al. [10], Attarha et al. [23], Hwang et al. [24], Ramdhani et al. [25], Boero et al. [26], Chakravarthy and Amutha [27] have used and modified the flow rules to improve the degree of load balancing but with increasing overhead and delay. Priyadarsini et al. [28] considered self-adaptive load balancing scheme (without considering delay) to minimize the switch migration but with the increase in cost and delay.
Towards an Efficient Resource Allocation based on Software-Defined Networking approach
2021, Computers and Electrical EngineeringCitation Excerpt :But in the event of congestion, the approach cannot meet the requirements of the real-time traffic. In [21], Boero et al. proposed a novel method based on the modification of the Beacon controller to guarantee QoS needs for the flows. In this method, the Beacon controller reroutes the flows among queues taking into consideration flow rates and packet loss in each queue to make the adaptation.
Machine intelligence approach: To solve load balancing problem with high quality of service performance for multi-controller based Software Defined Network
2021, Sustainable Computing: Informatics and SystemsA survey of low-latency transmission strategies in software defined networking
2021, Computer Science ReviewCitation Excerpt :Finally, the path with the highest probability is selected for traffic distribution. Boero et al. [139] have inserted an AQM mechanism into the rerouting process. First, the data plane information collected by the improved Beacon controller is used to schedule the flow on the congestion queue in the switch to the idle queue.
A comprehensive survey of load balancing techniques in software-defined network
2021, Journal of Network and Computer ApplicationsCitation Excerpt :Therefore, improving the efficiency and overall performance of the system. The QoS aims to provide a better user experience by preventing excessive device latency, optimizing performance, and reducing response time (Šeremet and Čaušević, 2020; Karakus and Durresi, 2017a; Boero et al., 2016). This is one of the overarching goals of LB since the proper use of resources is important for the efficiency of the SDN LB model.
Hierarchical Queue Management Priority and Balancing Based Method under the Interaction Prediction Principle
2023, Electronics (Switzerland)
Luca Boero was born in Genoa, Italy in 1989. In 2012 he got his Bachelor Degree in Telecommunication Engineering at the University of Genoa; in 2015 he achieved a Master Degree in Multimedia Signal Processing and Telecommunication Networks with a thesis on Quality of Service in Software Defined Networking. From November 2015 he is a Ph.D student at the University of Genoa. His main research activities concern networking and SDN.
Marco Cello was born in Savona, Italy in 1983. He got his “Laurea Magistrale” (M.Sc.) degree cum laude and his Ph.D in 2008 and 2012, respectively both at the University of Genoa. In 2012, 2014 and 2015 worked as Post-doc research fellow at University of Genoa with a fellowship funded by Fondazione Carige. In 2013 he was Post-Doc research fellow at Polytechnic Institute of New York University and Visiting Research Fellow at New York University Abu Dhabi. He is currently Post-Doc research fellow at Nokia Bell Labs in Dublin, Ireland. He is a researcher on networking with almost 8 years of experience, including managing research projects funded by national industries, the European Community and the European Space Agency (ESA). He has strong expertise in software for simulation, Linux-based emulation of telecommunication networks and Linux administration. He is a co-author of over 20 scientific works, including international journals, conferences and patents. His main research activities concern: Network Modelling/Teletraffic Engineering; Call Admission Control; Routing and Congestion Control in Delay Tolerant Networks; Software Defined Networking.
Chiara Garibotto was born in Chiavari, Italy in 1985. In 2012 she got her Bachelor Degree in Telecommunication Engineering at the University of Genoa; in 2015 she achieved a Master Degree in Multimedia Signal Processing and Telecommunication Networks with a thesis on Quality of Service in Software Defined Networks. From November 2015 she is a Ph.D student at the University of Genoa. Her main research activities concern networking and signal processing.
Mario Marchese was born in Genoa in 1967. He got his degree with honors from the University of Genoa in 1992 and the PhD in July 1997. From 1999 to 2005 he worked at the National Consortium for Telecommunications (CNIT). Associate Professor at the University of Genoa from 2005 to January 2016, since February 1, 2016 he has been Full Professor at the same University. He researched at the German Aerospace Center (DLR), as a Visiting Professor / Guest Scientist. He is the founder and head of the Laboratory “Satellite Communications and Networking”. He coordinated the technical-scientific and financial management of many research projects. He got 4 patents. He has published over 280 scientific papers including 1 international book, 2 edited books, 77 articles in international journals and 13 book chapters. He has been the Coordinator of the PhD in ”Science and Technology for Electronic and Telecommunication Engineering” since 2013 . He was “Chair” (2006–2008), “Vice-Chair” (2004–2006) and “Secretary” (2002–2004) of the “Satellite and Space Communications Technical Committee” of ”IEEE ComSoc”. He is the winner of the IEEE ComSoc “Satellite Communications Distinguished Service Award” in 2008 and of numerous “Best Paper Award”. His main research interests concern: space, satellite and heterogeneous telecommunications networks, quality of service over heterogeneous networks and applications for smartphones.
Maurizio Mongelli got his PhD degree in Electronic and Computer Engineering at the University of Genoa (UniGe) in 2004. His PhD was funded by Selex Communications SpA (Selex). He worked for both Selex and the CNIT from 2001 to 2010. During the PhD and in the subsequent years, he worked on quality of service for military networks for Selex. From 2007 to 2008, he coordinated a joint laboratory between UniGe and Selex, dedicated to Ethernet resilience. He was recently the CNIT technical coordinator of a research project concerning satellite emulator systems, funded by the European Space Agency; he spent 3 months working on the project at the German Aerospace Centre in Munich, Germany. He is now a researcher at the Institute of Electronics, Computer and Telecommunication Engineering of the Italian National Research Council. He is a co-author of over 70 scientific works, including international journals, conferences and patents. His main research activity concerns resource allocation and optimization algorithms for telecommunication, machine learning and cybersecurity.