eBooks

Solving NFV's Data Plane Paradox

Issue link: https://hub.radisys.com/i/859789

Contents of this Issue

Navigation

Page 3 of 7

and requirements compared to the control plane. A data plane function will require fewer CPU cycles per gigabit of traffic, but will be called on to transport, encrypt, and encode, millions of bits through the network – potentially scaling to 100s of Gbps if not multiple Tbps. A recent white paper (3) on the subject illustrated this issue well: analyzing a three minute VoIP call, the paper outlined that the control plane was required to process just 10-15 SIP signaling packets, whereas the data plane handled 36,000 RTP media packets. Scale that up to thousands of broadband streams, millions of VoIP calls, and then introduce video which is exponentially higher in terms of bit rates and you quickly get a sense of the packet handling requirements placed on the data plane within the network. To meet these demands in the network, operators are realizing that a different class of virtualization platform is required for data plane NFV; certainly at a different scale and reliability level compared to enterprise-class virtualization and cloud platforms. These differences center on the telecom industry's reliance on carrier grade – with the key requirements being reliability, low latency and ability to scale while operating at wire- speed. In a survey carried out by analyst firm Heavy Reading, the worry that NFV elements may not be "carrier grade" was rated as a key concern by nearly 40% of operators surveyed. And yet there is a paradox here, as the ability to use commercial servers, and not dedicated "telco" equipment, was also cited as a key driver for NFV. In other words, the key strength of NFV (commercial off the shelf hardware supporting virtualized functions) is also seen as a key potential weakness – the worry that these platforms will not be "carrier grade". SDN AND NFV Software Defined Networking splits up the functions of the current routers and switches in the network, taking the control layer intelligence and putting that in centralized controllers that then manage simplified IP packet forwarders. Now, data packets are forwarded by an optimized data plane, while the routing (control plane) function is running in a virtual machine enabling autonomous management of the network. Although NFV and SDN can exist independently, there are clear synergies between the two, as NFV can provide an infrastructure upon which SDN can operate. The original NFV white paper of 2012 described the relationship in these terms: "Network Functions Virtualization is highly complementary to Software Defined Networking (SDN), but not dependent on it (or vice-versa). Network Functions Virtualization can be implemented without a SDN being required, although the two concepts and solutions can be combined and potentially greater value accrued." CONTROL PLANE OPTIMISATION As the basic benefits have been proposed, and the enabling architecture defined, much of the early focus and early successes within NFV have been on the control plane elements in the network. This is the signaling infrastructure that establishes sessions, authorizes users, and sends control commands across the network. The reason for this is that these control plane elements tend to be CPU-intensive processes but are much lighter in terms of throughput and bandwidth requirements. Control plane functions are well suited to NFV architectures, and existing virtualized platforms. FOOTNOTES: 1. Network Functions Virtualization - Introductory White Paper http://portal.etsi.org/nfv/nfv_white_paper.pdf 2. ETSI NFV Latest Drafts:http://docbox.etsi.org/ISG/NFV/Open/Latest_Drafts/ Leading vendors have demonstrated prototype or deployed virtual EPC (evolved packet core) and IMS elements, as well as Cloud-RAN prototypes with virtual base station control functions running on pooled base band resources. Some mobile operators at the head of this curve have already virtualized smaller scale deployments of their core network functions in a datacentre environment. So we can see that, in simple terms, control plane functions tend to be CPU intensive, lending themselves well to an architecture that decouples that core resource from the actual function in hardware, and makes it available to be cycled up and down as demands are placed on the network. But although this control-plane focus does offer some "quick (or perhaps quicker) wins" for NFV, it risks ignoring a key aspect of the NFV-based network – the data plane. DATA PLANE AND NFV Whereas we defined the control plane as the controlling signaling infrastructure of the network, establishing and managing sessions end to end across the network, the data plane refers to the actual transport and delivery of the broadband traffic and media that are flowing within those sessions. Within an EPC, the data plane elements are the Serving Gateway (S-GW) and Packet Gateway (P-GW) that move user traffic from the user device to external networks. Other data plane Virtualized Network Functions (VNFs) include DPI elements, as well as policy-related elements (PCEF/ PCRF). IP Forwarding elements that are controlled by SDN controllers also reside in the data plane. This differing role of the data plane - the actual processing and delivery of traffic - means that the data plane has distinct scale A TMN eBOOK: SOLVING NFV'S DATA PLANE PARADOX

Articles in this issue

Links on this page

view archives of eBooks - Solving NFV's Data Plane Paradox