The system of Internet highways and byways is being reconfigured to cope with the size and shape of traffic heading over it, while Internet companies are dreaming up fresh approaches to avoid
It's an old story with a new twist. The big fret among users was once the idea that the Internet would run out of bandwidth before the end of the 2000s, or that its IP address space was soon to be exhausted. Not every techno-prophet subscribed to these notions, but most agreed that the somewhat haphazard fashion in which the Internet was built-out in the 1990s and 2000s did not take account of the traffic demands that it faces in the 2010s.
The IP address availability issue rumbles on still, as the remedy - migration from the legacy IPv4 protocol to IPv6 - is only just getting going in most regions. The core bandwidth issue - that is, whether the existing Internet infrastructure can manage with the proliferating volumes of traffic being loaded onto it - has largely been sorted out by advances in switch/router technology and ingenious innovations by the networking companies for obtaining greater capacity from the available routes.
This is despite the accelerating pace of growth in Internet traffic generated by video, and generally by the increasing number of large unstructured files shunted around. This is causing many Internet strategists to reconsider how the Internet of the future should be planned.
The Internet was, of course, constructed with traffic management in mind: routers constantly check the available routes to an IP packet's destination, and send it on its way via the route of least contention. They often send packets the long way round, where the shortest path appears congested with previously-sent packet traffic.
"Lack of raw bandwidth is unlikely to be a gating issue, especially as '100G' becomes the new currency in the backbone of IP and optical networks," says Houman Modarres, senior director of marketing at the IP Division of Alcatel-Lucent. Fibre networks existing or being deployed will have plenty of capacity to cope with further increases in traffic volume with the help of a technique called dense wave division multiplexing (DWDM), which enables single fibres to carry multiple channels, each encoded at a different wavelength, up to a maximum of 160 at 10Gbps each at present.
This adds up to an aggregate bit-rate of 1.6Tbps per fibre, and given that there are laboratory demonstrations of up to almost 1,000 40Gbps channels, this is set to increase another 25 times or so over the next decade. At the same time vendors are looking at stepping up from 100Gbps to 400Gbps over single-fibre channels.
The changing profile of data, with the growth in large unstructured files and online video, poses a new challenge of timely reliable delivery, irrespective of what theoretical bit-rate is available. The danger is of the Internet choking on these large files or video streaming sessions, and thereby failing to meet requirements for latency and'mandated Quality of Service. For live video streaming, delayed packets are as worthless as dropped ones, resulting in flaky picture quality with no scope for retransmission.
The changing nature of Internet traffic is creating issues not in capacity, but performance and ability of the core switching fabric to cope with the anticipated level of routing within the Internet. For this reason the major Internet infrastructure vendors such as Cisco Systems, Juniper Networks, and Alcatel-Lucent are still investing in their core routers and switches, and are keen to draw attention to the scale of the expected traffic deluge.
In February 2012 Cisco published its latest Global Visual Networking Index (VNI) focusing on the growing impact of mobile data, which it predicts will increase by 18'times between 2011 and 2016, with smartphone traffic rising by 50 times and tablet traffic by 62 times. The number of Internet-connected devices will soar to 50'billion by 2020 according to some estimates, as the 'Internet of Things' establishes itself (see box, p41). This gives rise to another issue: how to handle a vast number of small data transmissions, more particularly at the edge of the network, according to Dominic Elliot, solutions architect at Cisco.
"Edge devices are increasingly called on to deal with large amounts of small transmissions associated with signalling or data bursts," Elliot says. "It is essential that as we scale the network and service elements they are capable of meeting the changing nature of the Internet traffic profile." This will require increased CPU capability within the network, optimised to deal with these numerous small bursts of data.
In the immediate future the greater challenge lies in handling the growing movement of large unstructured data files between two or more sites, according to Steve Broadhead, director of vendor-independent testing organisation Broadband-Testing. Such files include video and high-resolution digital images, and these cause issues regardless of the bandwidth of the connection. In some cases there is no hurry to deliver the file, but in others, including streaming video, latency is a big issue.
According to Broadhead: "The problem here is that traditional data transfer protocols such as File Transfer Protocol (FTP) are designed neither to take advantage of large data pipes, nor to resolve latency issues. So, for usage such as data-centre-to-data-centre transfers, or in specialist industries such as medical and geophysics, where they need to transfer digital images as fast as possible between two points across the globe, there is a real problem to be solved."
WAN optimisation boost
As Broadhead also points out, this has given renewed impetus to the WAN optimisation business, with emerging vendors such as Talon Data and Bitspeed creating technology to optimise these large data transfers at connection speeds up to 10Gbps. The assumption now is that the bottleneck lies in the network's ability to deliver all the packets within an allotted time frame rather than the raw bandwidth itself. The focus is on judicious use of cache storage within the network, along with coupling between the source and destination, to minimise the extent of packet retransmissions. Talon Data Systems, for instance, uses a proprietary technique on top of the Internet's Transport Control Protocol (TCP), using a combination of buffering and interaction between the sender and receiver to keep careful track of packets dropped during transmission, and retransmitting as required as fast as possible.
Bit Speed's system, called Velocity, is an interesting case in that it splits file transfers into parallel streams to make full use of the network's full capacity, while TCP by itself transmits over a single path. This is similar in principle to the Multipath TCP developed by the Trilogy Project (see box, left) that may well become an integral part of the IP protocol stack supported right across the Internet, in due course.
IPv6 migration issues
The much bigger change regarding the IP protocol stack though is coming with migration from IPv4 to IPv6. IPv4 emerged as the Internet's first protocol in the 1970s, using a 32-bit addressing scheme, which meant that, in theory, it could only support about four billion connected devices. In anticipation that this address space would become exhausted as the Internet went public in the 1990s, IPv6 was developed, doubling the address space to 64 bits, which is far larger than will ever be required to support the 'webosphere', while bringing various efficient improvements that exploit the increased number of bits in the header to accelerate routing calculations.
However, techniques such as Network Address Translation (NAS) enabled the dwindling stock of IPv4 addresses to sustain operations for longer than expected, by allowing multiple computers and other IP devices to share one IP address. As a result, only about 1 per cent of Internet packets are IPv6, and only 0.15 per cent of the top million websites are accessible via the new protocol.
The acceleration in growth of IP-connected devices engendered by the Internet of Things and online video services has brought matters to a head, and from this year (2012) migration from IPv4 to IPv6 will gather pace.
IPv6 will help cope with the Internet traffic explosion, having itself evolved since its inception in the 1990s to incorporate better support for mobile data and video, with mechanisms for Quality of Service and a simple method of supporting roaming devices. It caches a device's home address alongside the temporary address as it roams, enabling IPv6 packets to be readily routed to the correct node. Under IPv4, this was bolted on, with varying implementations.
Complexity comes with cost
As Broadband-Testing's Steve Broadhead notes, while on the whole IPv6 will make routing more efficient, it does increase the immediate packet processing overhead since the header is bigger: "Carrying all this information will certainly put more strain on existing router/switch-based architectures that are IPv6 compliant, but were designed for use with IPv4." Although IPv4 and IPv6 are incompatible, leading IP switching vendors such as Brocade have come up with tools to help support migration and enable a degree of interoperability during the process.
"Our ServerIron ADX switches can accept IPv6 requests arriving from IPv6 clients, and translate them into IPv4 requests'for internal hosts that do not communicate in IPv6," notes Pavel Radda, marketing manager at Brocade. "They can also insert the original IPv6 client IP addresses so that IPv4 hosts can use that information when required." Given that migration could be a lengthy process, such ability to make use of IPv6 information could prove valuable. Eventually IPv6 will become predominant, and be supported by all hosts, routers and switches: then the migration issue will fade away. However, the need to continue improving the Internet's switching fabric will never go away for, as recent experience has shown, it has to cope not just with continual increases in traffic data volume, but also ever greater unpredictability and changes in the profile of the files and sessions that it needs to handle.
While the network will cope in terms of raw bandwidth, there is a danger that costs of managing the extra complexity could escalate, unless there is continuing innovation and integration of higher-level intelligence into core routing and switching products. "To avoid breaking the economics of core network provisioning, service providers must extract every bit of cost from their core networks without compromising services or reducing quality of experience for their end users," says David Noguer Bau, head of service provider marketing at Juniper Networks. "Success hinges on finding an economical, more scalable, and more efficient model for building and maintaining core transport networks."
This will require radical changes in approach, right down to the core ASICs (application specific integrated circuits) at the heart of the systems, according to Bau, who referred to Juniper's Junos Express chipset designed to address the challenges of scale as well as speed and cost.
Yet, while such developments are important, there is also, as Cisco's Elliot acknowledges, a shift in emphasis away from a focus on the capability of the underlying switching fabric towards intelligent routing of data at a higher content level, making optimal use of caching within the Internet to overcome bottlenecks and ensure timely delivery of IP packets. "Most providers are now striking a balance between careful capacity planning of their core and content optimisation techniques," says Elliot.
There is one thing though that is beyond the capability of any vendor to change, and that is the speed of light. This imposes a fundamental constraint over the ability of service providers to meet demands for low latency for real-time applications.
As Brocade's Radda agrees, service providers and vendors alike cannot alter this fundamental physical law, but can only do their best to accommodate it by distributing content in caches and ensuring that switching and routing functions add as little as possible to the overall delay.
"Carriers are addressing this by not only being geographically closer to the points they need to deliver data to, but through ultra low latency equipment," says Radda. Further progress by the vendors could shave useful microseconds over round trip latencies, but the major contribution will be made by higher level measures involving caching and intelligent content distribution. In order to maximise efficiency and minimise costs, more of the intelligence required to manage content distribution will be embedded into the network infrastructure.