ABSTRACTS OF ARTICLES OF THE JOURNAL "INFORMATION TECHNOLOGIES".
No. 4. Vol. 25. 2019

DOI: 10.17587/it.25.223-228

I. V. Philippov, Software Engineer, ili.filippov@gmail.com, A. F. Melik-Adamyan, Software Architect, areg.melik-adamyan@intel.com, V. A. Sukhomlinov, Software Engineer, vadim.sukhomlinov@gmail.com, Intel Corporation US, USA

Cloud Deployment of Network Functions

The article discusses the problem of the deployment of network functions to a cloud infrastructure. We propose network function representation as a microservice and introduce a notion of the "Cloud Boundary Node" (CBN) for the boundary between functions system and another network. CBN classifies each incoming packet, determines a service chain for it, encapsulates packet to internal representation and sends it for processing, taking load balancing into account. Service chains are represented as a packet processing graph that should be implemented in a "run-to-completion" methodology — whole service chain is done at one server using shared memory technology. Packet processing graph can be built using NFF- GO system, which provides efficient automatic intra-node scaling. Graph structure is the same for all the nodes; however, the graph can be extended by using plugins, which are determined at the packet classification stage.
As a result, our architecture is easier to manage due to only one single management object — CBN, where the administrator can set the required service chains for every packet. Architecture can be better scaled and effectively handle various failures because all servers have the same processing graph, and CBN can simply rebalance traffic between them. Great performance is achieved by eliminating the costs of data transferring and commutation: only one commutation happens while CBN classification stage and only one data transfer happen between CBN and handling server.
Keywords: Network functions, Cloud infrastructure, Microservices, Virtualization, Containerization, NFF-GO

To the contents