Cell-phone subscriptions are reaching saturation causing revenue from mobile voice services to plateau. However, mobile-data revenue is still in the early-adopter phase and is ramping up. Figure 1 below shows IDC estimates for mobile voice and data revenue growth and illustrates that voice revenue growth is stalling while there was good revenue growth for mobile broadband data service revenue between 2008 and 2009 and significantly more revenue growth predicted for 2010 and beyond.
Figure 1: IDC predicts little mobile voice services revenue growth but good growth for mobile broadband data services revenue
Consequently, the communication industry recognizes that mobile broadband services will be the engine for future revenue growth and the industry is now developing suitable infrastructure to better handle mobile broadband services. Existing 3G (third-generation) networks are increasingly strained from the increased data traffic so operators worldwide have announced plans to move from 3G to higher-speed LTE (long-term-evolution) networks, which deliver data rates in excess of 100 Mbps and promise a fivefold increase over 3G HSPA+ (high-speed-packet-access) networks. However, these network upgrades alone won't deliver sufficient data throughput to guarantee good service to all users based on real customer-traffic patterns.
Bandwidth improvement alone is not sufficient. Additional technologies such as DPI (deep packet inspection) will be required to ensure that prioritized and managed traffic optimizes the mobile user’s experience. DPI expands effective network capacity by managing and optimizing data traffic. It goes far beyond the simple examination of IP (Internet Protocol) headers to determine traffic routing by examining packet contents to determine the actual use for each packet. All packets are not equal; Is it an e-mail, Web, video, or P2P (peer-to-peer) packet? Some packets (such as video packets) demand shorter delivery latency while user experience suffers little from short delays in email and HTML Web packet delivery.
Developers of the 3G and LTE networks employ a single shared data channel for all subscribers in a given cell. Carriers assumed that users would employ mobile data primarily for “bursty” activities such as Web surfing and e-mail when these networks were first designed. A shared data channel provides the same high-bandwidth pipe for video downloads, voice, e-mail, and for static HTML Web page traffic. Shared data channels perform poorly for the large sustained transfers of streamed data such as video.
High-bandwidth streams overfill the communication channel beyond capacity, resulting in dropped packets and long latencies for all cellular traffic. The growth in mobile data use and broad adoption of video and P2P exchanges—with their large sustained transfer requirements—are increasingly common. According to a forecast by Cisco (see Figure 2), video and P2P traffic currently accounts for 60\% of all data and will grow rapidly.
Figure 2: Cisco predicts that video service traffic will continue to grow rapidly.
In the early days of the Internet, IP headers clearly designated the target application but most traffic today simply looks like Web traffic based on header information. By using DPI, networks can develop a better understanding of how customers are using the service and can deliver packets using the right QOS (quality of service) criteria for better end-user experience.
DPI technology is simple in concept but complex in practice. Conceptually, inspecting a packet to determine subscriber and application type and then acting on that information looks easy. However, network line rates and rapidly evolving applications add complexity. Based on present data rates, packet rates are already staggering. One 10-GbE (gigabit-Ethernet) channel can carry 30 million minimum-sized packets/sec. Even with a realistic traffic profile and 200-byte packets, a 10-GbE channel carries 10 million packets/sec. At that speed, there’s only 100 nsec to receive and inspect each packet, determine its application, modify it if necessary, and forward it to the next node in the network. A 3-GHz, single-core processor can only execute 300 instructions in that amount of time, which is not enough to receive the packet let alone inspect and process it. This harsh reality is driving the rapid adoption of multicore, multithreaded processors for packet inspection. A multicore processor running at a modest 1-GHz clock rate can handle the work load by using multiple threads to attack the problem.
ATCA board vendors are starting to release multicore server blades for LTE network development that incorporate multiple Intel® Xeon® multicore embedded processors based on the Nehalem architecture. For example, Continuous Computing’s multicore XE60 Dual Nehalem Blade (see Figure 3) sockets two high-performance Intel® Xeon® 5500 series processors, permitting the board to carry as many as eight processor cores in two quad-core processors. The XE60 also accommodates as much as 64Gbytes of memory in eight DDR3 DIMM sockets and as many as four hard drives directly attached through on-board SAS/SATA interfaces.
Figure 3: Continuous Computing’s XE60 Dual Nehalem Blade supports as many as eight processor cores and 16 threads
Figure 4 shows a block diagram of GE Fanuc’s GEFIP A10200 ATCA Dual Nehalem Blade single-board computer. It too combines two Intel® Xeon® 5500 series processors for as many as 16 threads running on eight on-board cores using two quad-core processors. Each processor has direct access to four DDR3 DIMM sockets that can hold as much as 64Gbytes of DRAM in total.
Figure 4: GE Fanuc GEFIP A10200 Dual Nehalem ATCA Single-Board Computer
By using DPI, network operators can create and administer service plans with different delivery terms and rate structures that will attract a wider subscriber base. Operators can optimize some low-data-rate service plans for Web surfing and e-mail sessions using only 64-kbps channels with a tight bandwidth cap on faster traffic. Another service plan might offer 250-kbps, YouTube-style video streaming but limit high-definition, multi-megabit video streaming. Yet another service plan might be designed to attract multiplayer-game enthusiasts, offering low latency for gaming packets. Corporations might choose a premium service package with traffic priority for e-mail, CRM (customer-relationship management), and other corporate applications. A service for P2P users could offer unlimited bandwidth during off-peak hours but tightly cap P2P bandwidth during peak-usage times. By offering diverse plans and fee structures, network operators can manage network traffic while attracting more customers, which will accelerate the subscriber and ARPU (average revenue per user) growth they perpetually seek.
Are you designing systems for LTE networks or any other applications that make good use of multicore processing? What have you discovered about system design with multicore processors?
Note: Continuous Computing and GE Fanuc are Associate members of the Intel® Embedded Alliance.
Mike Coward, Deep Packet Inspection Optimizes Mobile Applications, Continuous Computing, http://www.ccpu.com/papers/dpi-mobile-apps/
Roving Reporter (Intel Contractor)
Intel® Embedded Alliance