Optimize the Cost-Performance of LTE Networking Equipment
To achieve the necessary wire-speed performance for large numbers of virtual networks, the underlying software architecture must provide optimized support for key technologies such as virtualization, SDN and DPI.By Charlie Ashton, 6WIND
LTE is commonly viewed as the essential enabler for meeting ever-increasing user demands for mobile data bandwidth. Several forecasts, including the Cisco Visual Networking Index Forecast, (http://www.cisco.com/en/US/netsol/ns827/networking_solutions_sub_solution.html ) predict mobile broadband growth of 18X over the next five years. Driven by the tremendous increase in data-enabled devices such as smartphones, tablets, machine-to-machine devices and netbooks, network bandwidth is being consumed at a pace that service providers are challenged to sustain. As the use of video continues to explode – the same forecast estimates that by 2016, 71 percent of all mobile data traffic will be video – and the move to cloud-based services accelerates, there is no end in sight to users’ voracious appetite for data.
To manage all of this traffic, offer new services and more effectively monetize their networks, service providers are developing applications that utilize network intelligence technologies. Support for software-defined networking (SDN) and deep packet inspection (DPI) technologies provides more flexible, efficient networking and the ability to offer more advanced, content-based services, including security and bandwidth management applications such as policy and charging control, quality of service (QoS), subscriber analytics and traffic optimization. DPI is also a fundamental technology for services which require policy-driven, real-time charging and content distribution. Critical to the success of these technologies is having sufficient computing resources to execute these DPI and data-driven applications without a significant increase in networking equipment cost or power consumption.
All of this leads to an interesting question: Are more powerful processors and faster I/O hardware alone enough to meet these bandwidth and performance demands?
In a word – no. Hardware alone cannot meet these requirements, at least not at a price service operators can afford. What is also needed is efficient software, and in particular, packet processing software.
Software – The Key to Cost-Effective, High-Performance Networks
To achieve the necessary performance levels associated with the move to 4G/LTE technology, equipment providers have been moving to software architectures optimized for packet processing. These architectures take advantage of the fact that in a typical 4G networking environment, over 90 percent of the workload is data-plane packet processing and forwarding. With this workload profile, performance is limited by the overheads and latencies inherent in standard operating system networking stacks.
|Figure 1: Typical 4G Workloads are only 10% control plane.|
Architectures optimized for packet processing split the networking stack into two layers. The lower layer, typically called the fast path, processes the majority of incoming packets on dedicated CPU cores outside the OS environment and without incurring any of the OS overhead that degrades performance. Only those rare packets that require complex processing are forwarded to a Linux networking stack, which performs the necessary management, signaling, and control functions.
The fast path architecture exhibits linear performance scalability until the platform limits are reached. One benchmark using the 6WINDGate software from 6WIND achieved a 10x network capacity improvement versus the standard operating system stack on an Intel® Xeon® Processor E5-2600 Series platform.
Extensions for Cloud Computing
Virtualization support requires that the packet-processing software run as part of the virtual appliance or network, thereby providing its services transparently to the higher-level applications. Compatibility with standard hypervisors is also a requirement. Advanced architectures make use of several techniques to maximize system performance by removing key I/O bottlenecks: virtual NIC (vNIC) drivers, direct VM-to-VM communication, and I/O virtualization (IOv). The vNIC driver leverages communications between VMs via the virtual switch, making the development and provisioning of systems with multiple VMs more efficient. For higher system performance, the VM-to-VM driver allows inter-virtual-machine communications to bypass the hypervisor’s virtual switch, while IOv removes the virtual NIC emulation and allows direct access between the physical NIC and the bottom of the network stack.
|Figure 2: 6WINDGate implements an IOv direct connection to the fast path and this add-on to the base package provides support for multiple industry-standard, IOv-enabled network interface cards (NICs) and removes the performance penalty imposed by the virtual switch.|
Key to software-defined networks is the ability to dynamically allocate resources to the ever-shifting requirements of the network. High-performance packet-processing architectures utilize a dedicated pool of cores that can be reconfigured dynamically to run either the control plane or data plane in line with network parameters. Resources can be allocated and de-allocated to match traffic requirements, providing optimum network monetization. Advanced architectures use a hybrid design that enables either the local control plane or SDN products such as OpenFlow to manage the flow table and associated virtual routing and forwarding.
DPI performance improvement can be achieved by placing the DPI flow table within the fast path. By triggering the DPI engine only in the cases of relevant packets or flows, while implementing a smarter mechanism for allocating packets and flows to specific cores, the system-level performance is maximized while processing the packets with zero loss. Performance can increase up to 7x through this approach. The overall efficiency of the platform is maximized by ensuring that only relevant packets are sent to the DPI engine for full processing and that the DPI engine is bypassed in all other cases.
The types of packets that are sent to the DPI engine include:
- Non-empty packets – for example a pure signaling ACK does not need DPI processing
- Or the first packets of new flows are sent to be classified
- Or packets that require detailed analysis – for example SIP packets or ftp packets
- Or packets from flows that may need to be reclassified because of the specifics of the application – for example security applications where the flows have to be analyzed continually to detect any new URLs that are requested
The Data-Driven Future
A high-performance networking infrastructure is essential to the success of the mobile network operators’ (MNOs’) move to the network-as-a-service (NaaS) business model. To achieve the necessary wire-speed performance for large numbers of virtual networks, the underlying software architecture must also provide optimized support for key technologies such as virtualization, SDN and DPI. The availability of a high-performance packet-processing foundation enables this move and is key to the future success of mobile computing.
Charlie Ashton is VP of marketing and business development at 6WIND and is responsible for 6WIND’s global marketing initiatives and partnerships worldwide with semiconductor companies, subsystem providers and embedded software companies. Charlie has extensive experience in the embedded systems industry, with his career including leadership roles in both engineering and marketing at software, semiconductor and systems companies. He led the introduction of new products and the development of new business at Green Hills Software, Timesys, Motorola (now Freescale), AMCC, AMD and Dell.