Attila Takacs

Scalable fault management for OpenFlow

In the OpenFlow based split architecture, data-plane forwarding is separated from control and management functions. Forwarding elements make only simple forwarding decisions based on flow table entries populated by the controller. While OpenFlow does not specify how topology monitoring is performed, the centralized controller can use Link-Layer Discovery Protocol (LLDP) messages to discover link and node failures and trigger restoration actions. This monitoring and recovery model has serious scalability limitations because the controller has to be involved in the processing of all of the LLDP monitoring messages. For fast recovery, monitoring messages must be sent with millisecond interval over each link in the network. This poses a significant load on the controller. In this paper we propose to implement a monitoring function on OpenFlow switches, which can emit monitoring messages without posing a processing load on the controller. We describe how the OpenFlow 1.1 protocol should be extended to support the monitoring function. Our experimental results show that data plane fault recovery can be achieved in a scalable way within 50 milliseconds using this function.

Network Function Placement for NFV Chaining in Packet/Optical Datacenters

In an operators datacenter, optical technologies can be employed to perform network function (NF) chaining for larger aggregated flows in parallel with the conventional packet-based fine-grained traffic steering schemes. When network function virtualization (NFV) is enabled, virtualized NFs (vNF) can be placed when and where needed. In this study, we identify the possibility of minimizing the expensive optical/electronic/optical (O/E/O) conversions for NFV chaining in packet/optical datacenters, which is introduced by the on-demand placement of vNFs. When the vNFs of the same NF chain are properly grouped into fewer pods, traffic flows can avoid unnecessary traversals in the optical domain. We formulate the problem of optimal vNF placement in binary integer programming (BIP), and propose an alternative efficient heuristic algorithm to solve this problem. Evaluation results show that our algorithm can achieve near-optimal O/E/O conversions comparable to BIP. We also demonstrate the effectiveness of our algorithm under various scenarios, with comparison to a simple first-fit algorithm.

Resource Management in Diffserv (RMD): A Functionality and Performance Behavior Overview

The flexibility and the wide deployment of IP technologies have driven the development of IP-based solutions for wireless networks, like IP-based Radio Access Networks (RAN). These networks have different characteristics when compared to traditional IP networks, imposing very strict requirements on Quality of Service (QoS) solutions, such as fast dynamic resource reservation, simplicity, scalability, low cost, severe congestion handling and easy implementation. A new QoS framework, called Resource Management in Differentiated Services (RMD), aims to satisfy these requirements. RMD has been introduced in recent publications. It extends the IETF Differentiated Services (Diffserv) architecture with new admission control and resource reservation concepts in a scalable way. This paper gives an overview of the RMD functionality and its performance behavior. Furthermore, it shows that the mean processing delay of RMD signaling reservation messages is more than 1330 times smaller then the mean processing delay of RSVP signaling reservation messages.

Optical service chaining for network function virtualization

This article presents an efficient optical service chaining architecture for network function virtualization in data centers. Service chaining (i.e., steering traffic through a sequence of network functions) is one emerging application of software-defined networking. However, existing schemes steer traffic solely in the packet domain, which is well suited for fine-grained (e.g., peruser level) flows carrying a relatively small volume of traffic. This article discusses how packet-based schemes do not yield sufficient efficiency for large/aggregated flows steered through high-capacity core network functions. It introduces an optical steering domain into the operators data centers for NFV service chaining at a coarse-grained traffic level using wavelength switching. Performance evaluation shows that the optical steering domain can achieve significant power savings compared to using packet technologies as flow rates and the number of vNFs per service chain grow.

GMPLS controlled ethernet: an emerging packet-oriented transport technology

Most of the worlds traffic is now packet-based. While SONET/SDH transport technology continues to evolve, there are now major standardization efforts to develop a native packet-oriented transport technology. Ethernet, long dominant in the enterprise, is one of the fastest developing technologies for the transport layer. GMPLS, based on mature signaling and routing protocols, is gaining traction as a transport control plane, providing fast restoration and supporting automation of provisioning. We will briefly highlight recent Ethernet standardization and then introduce the concepts and standards work allowing GMPLS to control Ethernet.

Severe Congestion Handling with Resource Management in Diffserv on Demand

Quality of Service (QoS) for the Internet has been discussed for a long time without any major breakthrough. There are several reasons, the main one being the lack of a scalable, simple, fast and low cost QoS solution. A new QoS-framework, called resource management in differentiated services (RMD), aims to correct this situation. This framework has been published in recent papers and is extending the IETF differentiated services (diffserv) architecture with new admission control and resource reservation concepts in a scalable way. This paper focuses on proposing and investigating two resource reservation solutions on the problem of severe congestion situation within a diffserv-aware network utilizing an admission control scheme called Resource Mananagement in Diffserv (RMD). The different severe congestion solutions are compared using extensive simulation experiments.

Comparative Performance Analysis of RSVP and RMD

Service evolution towards QoS capable applications requires efficient resource reservation protocols. Currently, RSVP is a widely known protocol for this functionality in IntServ networks. Unfortunately, the processing power needs of RSVP make it to a less favoured candidate in high-speed environments. In these scenarios low-complexity DiffServ solutions have a clear advantage. A currently studied and industrially supported DiffServ conform resource management protocol is RMD. In this paper, we certify RMD as a simple and efficient protocol for unicast traffic by comparing the performance of RMD with RSVP, and also verify that the reduction of complexity does not entail loss in performance.

A practical method for the efficient resolution of congestion in an on-path reduced-state signalling environment

Currently, the standardisation of on-path signalling protocols is going on within the Next Steps in Signalling (NSIS) Working Group of the IETF. NSIS is responsible for the definition of a general IP signalling protocol. The first use case of the proposed protocol is flow-level resource management. One of the considered reservation methods, reduced-state mode, is based on the Resource Management in DiffServ (RMD) framework. Since it relies only on per-class state information in interior routers, it has a number of benefits including scalability, low complexity, and low memory consumption. However, the price of simplicity is decreased efficiency in case of exceptional situations. The most demanding task for RMD is the handling of congestion that may occur after a failure resulting in re-routing of flows onto a new path. Resolving a suddenly evolved overload without per-flow states is a highly non-trivial task. We present a low complexity mechanism which easily handles the undesirable situation, and we give guidelines to set the parameters of our scheme based on worst-case calculations.

Network function placement for NFV chaining in packet/optical data centers

Intelligent network function (NF) placement can minimize optical/electrical/optical (O/E/O) conversions for NF chaining in packet/optical data centers hosting NF virtualization (NFV). We design an efficient algorithm for NF placement, and show it can achieve near-minimum O/E/O conversions.

Resilient reduced-state resource reservation

Due to the strict requirements of emerging applications, per-flow admission control is gaining increasing importance. One way to implement per-flow admission control is using an on-path resource reservation protocol, where the admission decision is made hop-by-hop after a new flow request arrives at the network boundary. The next-steps in signaling (NSIS) working group of the Internet engineering task force (IETF) is standardising such an on-path signaling protocol. One of the reservation methods considered by NSIS is reduced-state mode, which, suiting the differentiated service (DiffServ) concept, only allows per-class states in interior nodes of a domain. Although there are clear benefits of not dealing with per-flow states in interior nodes — like scalability and low complexity —, without per-flow states the handling of re-routed flows, e.g., after a failure, is a demanding and highly non-trivial task. To be applied in carrier-grade networks, the protocol needs to be resilient in this situation. In this article, we will explain the consequences of a route failover to resource reservation protocols: Severe congestion and incorrect admission decisions due to outdated reservation states. We will set requirements that handling solutions need to fulfill, and we propose extensions to reduced-state protocols accordingly. We show with a set of simulated scenarios that with the given solutions reduced-state protocols can handle re-routed flows practically as fast and robust as stateful protocols.