| Title | : | Providing Resiliency for Service Function Chaining in NFV systems using a SDN-based approach |
| Speaker | : | Karthik Karra (IITM) |
| Details | : | Mon, 27 Aug, 2018 4:00 PM @ A M Turing Hall |
| Abstract: | : | In current networks, middleboxes such as firewalls, Network Address
Translators (NAT), Load Balancers have become an integral part of the
network’s forwarding infrastructure. Two new networking paradigms,
Software Defined Networking (SDN) and Network Function Virtualization
(NFV), have changed the way the network is handled and operated. In
NFV-based systems, network functions can be written as a software
module and deployed on a regular commodity server instead of a
dedicated hardware. Thus, these specialized and expensive middleboxes
can be replaced by such virtualized network functions (VNF) running on
servers. Each network packet is processed by several network elements including switches, routers, and VNFs, in a strict logical chain order called a Service Function Chain (SFC). Resiliency is an important aspect to be considered when deploying SFCs and is typically achieved using redundancy that can be expensive. In this work, we propose an architecture for SFC resiliency without redundancy, based on SDN controller and NFV concepts. The architecture consists of three components viz. SDN Controller, Storage Server and NFV-enabled server nodes. One of the main tasks of the controller is to periodically monitor the working of the NFV-enabled nodes. If the controller detects an NFV-node failure, all the network functions of the affected SFCs will be moved to other node(s) or server(s) on alternate network paths. To improve the overall resiliency of an SFC the Tabu Search algorithm is used to increase the minimum Mean Time To Failure (MTTF) value of an SFC. The proposed architecture has been modeled using a standalone Java-based discrete-event simulation environment. The results show that it takes approximately 600 to 3400 microseconds to recover from up to 9 component failures in an SFC. It is observed that the Tabu Search always increases the minimum MTTF value of an SFC irrespective of its length from its default minimum MTTF value obtained from K-shortest path. It is also observed that as the number of SFCs and length of the SFCs increases, there is a increase in the overall MTTF values by which the SFC becomes robust. A Proof-of-Concept (PoC) is being implemented in a Linux environment to demonstrate feasibility. The Floodlight SDN controller and IncludeOS based unikernels (these have the least boot time of all the virtualization platforms) acting as NFV-enabled nodes are implemented. In summary, the work shows that resiliency can be provided without redundancy, resulting in cost savings and higher revenues for the network service operators. |
