Modeling and analysis of robust service composition for network functions virtualization

Abstract

Fault tolerance is critical for constructing a reliable service in Network Functions Virtualization (NFV). In this paper, we propose novel models and algorithms that provide the resilience of NFV services from multiple node and link failures. We first design an optimization model and the PAR protection algorithm that can efficiently protect an NFV service demand from network failures without any action from a controller due to the diversity of flow assignment. We then develop an optimization model for total demand protection with a guarantee of recovering the whole demand volume. Further, a new restoration algorithm, namely UNIT, is proposed for the design of large survivable NFV-based networks with the recovery of the affected bandwidth under the uncertainty of multiple network failures. We analytically prove the performance guarantee of UNIT in comparison with the optimal static solution. The results of our experimental study in a Mininet-based environment with the Ryu controller show that a combination of PAR and UNIT efficiently protects NFV-based networks from failures in terms of both resource restoration and recovery time.

Introduction

To meet the ever increasing demands on new services, network service providers must continuously install, operate and perform maintenance on new physical equipments. It results in significant drawbacks on the launch of new services in terms of investment and operation costs. Network Functions Virtualization (NFV) is a promising approach to network architectures that enables the providers to deploy and manage new network services with agility. In an NFV-capable system, network functions (e.g., routing, firewall, deep-packet inspection) are migrated from traditional hardware based appliances to a virtualized pool of resources deployed in the cloud. NFV allows network service providers to dynamically consolidate an ordered list of virtual network functions (VNF), called service function chaining (SFC), for providing a network service in a fast and flexible manner. In such agile systems, it is critical to ensure service continuity, rather than focusing on platform availability. In particular, node or link failures should not result in service disruption. Gill et al. show that link failures that occur frequently and are disruptive in datacenters can decrease the volume of traffic delivered by up to 40% [1]. In addition, due to strict performance requirements of network functions, a short time interruption in these systems may cause severe degradation in network service quality. Therefore, a desirable goal is to design mechanisms of service composition for NFV that can be resilient to an inevitable reality of failures [2].

To ensure network resilience upon network failures, the spare capacity must be pre-allocated in the network. The design problem of a resilient network is to optimize the spare capacity and flow assignment for a given survivability requirement. Approaches used in survivable networks can be roughly classified into two categories: protection and restoration. A protection mechanism pre-calculates backup paths before a failure occurs while restoration refers to mechanisms that start to search for alternate paths after a failure. A restoration mechanism provides a better solution in terms of resource consumption. However, it requires more sophisticated signalling schemes, and hence the recovery time is higher. In this paper, we consider a combination of protection and restoration for the robust service composition in NFV-enable networks. The first question is to develop an efficient protection mechanism under which a service demand can rapidly survive on node or link failures without any action from the controller. For a critical service, an NFV provider needs to design a solution that can ensure that the whole bandwidth volume of a service demand is able to be restored. Hence, it raises a challenging question of finding the optimal spare capacity for protecting the whole demand volume from failures. Finally, in order to provide a practical restoration solution, it is necessary to study efficient use of shared capacity for restoration, taking into account the uncertainty of multiple node and link failures in the NFV system. We address such questions as an important part of a fault tolerant network design.

The major contributions of this paper are as follows:

• We develop an optimization model for an efficient protection mechanism based on the concept of partially disjoint paths in the flow assignment for a service demand in NFV with service function chaining. In case of failures, an amount of bandwidth required by a demand will survive without any action from a network controller due to the diversity of flow assignment. We propose a novel algorithm, namely PAR, which provides an efficient solution of the protection mechanism for a large-scale NFV network.

• We extend our optimization model to investigate the optimal spare capacity for 100% demand protection with a guarantee of recovering the total demand volume of services. It repairs individual flows by using both shared spare capacity and capacity released by the failed flows. The computation of optimal spare capacity is not limited to single node or single link failures.

• We propose a novel restoration algorithm, namely UNIT, to provide an online solution of restoration with respect to multiple link and node failures that arrive one-by-one without knowledge of future arrivals in a large-scale NFV network. The performance guarantee of UNIT is analytically proved in comparison with the optimal static solution.

• We complement our theoretical analysis with numerical results and experiments in Mininet [3] with the Ryu controller [4]. We observe that our protection approach based on partially disjoint paths outperforms the traditional protection approach based on fully disjoint paths. The results under both generated network topologies and real world datasets show that a combination of PAR and UNIT provides fast restoration of service after link and node failures.

In summary, we develop both optimization models and efficient algorithms for NFV resilience upon multiple node and link failures. The effectiveness of the proposed schemes is demonstrated by theoretical analyses and extensive simulations. Our models and algorithms enable service providers to quickly restore the whole affected traffic upon network failures occurred along a service path, which is particularly important in NFV. To the best of our knowledge, this is the first paper that provides a systematic study, with proven guarantees, to the problem of robust service composition for NFV with regard to the uncertainty of failures. In addition, the paper tackles several practical factors on the optimal resource allocation for protection and restoration including multiple link and node failures, the dynamics of a link metric system, and the considerations of a fairness criterion.

The rest of this paper is organized as follows. Section 2 reviews the related work. Section 3 describes the model and the problem of designing service composition in NFV that is robust to network failures. Section 4 presents our optimization model (PT-O) and algorithm (PAR) for the problem of protection using partially disjoint paths. We provide our optimization model (RT-O) and algorithm (UNIT) for the problem of restoration under the uncertainty of failures in Section 5. Section 6 presents the evaluation of our models and algorithms upon network failures in both generated network topology and real datasets. We conclude the paper in Section 7.