Jonathan Homa examines the benefits of taking a more holistic view of network optimisation
Imagine you are in charge of your country’s public transportation. You have three divisions, busses, trains, and aeroplanes. Each division shares patterns of passenger traffic with each other, but otherwise operate independently. The population is generally satisfied with the level of service, even when switching between modes of transport, but you know that this is primarily due to the fact that your budgets have been generous enough to allow each division to build sufficient route capacity and frequency.
You are now informed that, sorry, your budget is being cut by 20 per cent, and by the way, you better keep the same level of customer satisfaction because there is an election next year. You know that if you cut services across the board by 20 per cent that things will quickly fall apart. What do you do?
After consulting with operational planners, economists, and other savants, you decide to build a sophisticated model and develop accompanying algorithms that looks at the totality of passenger traffic and all three modes of transport, simultaneously. The fortunate result, you discover, is that you can provide travellers essentially the same level of service. While you need to give up excess capacity (there won’t be many open seats), by carefully planning routes and schedules among all three divisions, you can get people to their destinations in line with their current experience.
Moreover, you can translate the algorithms underlying the new scheduling into apps that enable travellers to plan their routes with higher precision, and decide among options with different transit times and costs. And if a problem develops with a route, then the application notifies travellers in real-time and immediately suggests alternatives.
This is the essence of multilayer optimisation. To understand why this is so urgently needed today for telecommunications networks, we need to turn back the clock.
Our predicament: parallel evolution
A scant 25 years ago, telecommunications networks were planned and optimised for fixed-rate voice traffic. Networks employed SDH/SONET technologies which mapped this traffic onto optical channels. In essence, we had a single layer network.
Then along came data services traffic for computer networking and the budding world wide web. This was packet traffic based on statistical multiplexing principles. The packet engineers built their own overlay networks of Layer 3 Internet Protocol routers and Layer 2 Ethernet switches, but always came knocking on the door of the established and robust SDH/SONET networks to hitch a ride for transport.
SDH/SONET didn’t mind so long as the packet traffic sat, figuratively, at the back of the bus and didn’t complain.
Over time, however, the strength of packet grew and the two worlds of fixed-rate ‘bell heads’ and statistical multiplexing ‘data heads’ evolved in wary co-existence. SDH/SONET transport was displaced by packet-friendly OTN, which was also needed to support faster optical line rates. Transport gear also began incorporating some Layer 2 packet-switching functionality, creating a new breed of packet-optical transport equipment.
Today data services and packet traffic are dominant, so you can say that packet has won. But it has been somewhat of a Pyrrhic victory because service providers are still independently planning, optimising, and operating two sets of networks – a packet network and a transport network.
Returning to the public transportation analogy, this status quo has been generally acceptable until now, since each network was able to build excess capacity so that end-customers received adequate service. But this situation is no longer sustainable for two reasons. The first is budgetary, as service providers need to reduce their capital outlays. The second is competition. Service providers need to streamline and speed up service delivery to maintain pace with the expectations being set by cloud-based services delivery. As we explore below, both these challenges can be met through multilayer optimisation and control of packet and optical transport networks.
How it works today
Before discussing optimisation, let’s review how networks are designed today. To make things simpler we’ll just look at a Layer 3 router network for data services, and an underlying Layer 0–1 optical transport network. In the real world, other services like voice and Fibre Channel must also be considered, as well as use of Layer 2 packet switching.
It starts with data services traffic. This traffic originates from and terminates to many types of end users: enterprises, small and medium businesses, mobile base stations and mobile exchanges, residential traffic aggregation switches, all manner of large and small data centres, and traffic exchange points with other router networks. All these streams of data traffic get built up into a holistic geographic traffic matrix, which is the foundational requirement for a network design. Other key considerations are service-level agreements (SLAs) specifying the mix between committed and excess information rates to any given user, service availability which governs network diversity, backup paths and cost targets.
With these inputs we can design the packet network layer, which is, essentially, deciding where to physically locate IP routers and how these should be connected with each other. Each router is specified in terms of capacity, the number of ports, and the speed of each port, typically 10, 25 or 100 Gigabit Ethernet (GbE). Internal routing tables are configured so that when packets arrive at one port, the router can decide which port the packets should exit. Routing rules will account for various traffic load conditions, and also for extreme circumstances such as when a link or port fails.
The actual connections among the routers are performed by the optical transport layer. This has multiple degrees of freedom: mapping the GbE router port interfaces into fixed (ODUn) or flexible (ODUflex) Layer 1 OTN channels, employing OTN switching for grooming efficiency and flexibility on top of OTN transport, aggregating OTN channels onto Layer 0 WDM optical wavelengths, routing the wavelengths themselves between the optical nodes, and the use of ROADMs for handling wavelengths passing through a node to eliminate the need for expensive electrical conversion and provide optical layer flexibility.
There are several noteworthy points with the current approach. In the Layer 3 packet network, routers know with whom they are connected, but have no idea how they are connected, and whether this is being performed efficiently or not. The Layer 0-1 optical transport network essentially just pairs up sets of interfaces across geographic distances, but has no knowledge of the services being carried.
As a result, while the current approach based on a ‘separation of concerns’ has a certain engineering elegance, the lack of knowledge that the packet and transport layers have of each other can clearly be improved upon by looking at them holistically.
Packet and optical transport networks are inherently agile, equipped with points of flexibility that can be controlled remotely through software. So, it would make sense that the major tool to harness this agility, to realise multilayer networking, is software-defined networking. SDN provides intelligent, centralised, real-time control over networks with application-level awareness of the services they support. SDN for the WAN is based on a hierarchy. Domain controllers gather information and extend real-time control over different layers or geographical clusters of networking equipment. For instance there will be packet domain controllers and optical domain controllers. In turn, these domain controllers report into a parent controller, or SDN orchestrator (SDN-O), which is the mastermind ‘orchestrating’ all the underlying network resources. The SDN-O is responsible for implementing service level requests coming from an operations support system, and continuously ensuring that all layers of the network are working in harmony with maximum efficiency. The benefits from an SDN-controlled multilayer approach fall into three categories: optimisation, restoration, and provisioning.
Multilayer optimisation continuously reorganises Layer 0-1-2-3 network elements to handle both existing and incremental new service requirements in the most efficient manner, delaying the need to add new resources for new connectivity requests. Let’s look at a simple example. Assume that router A connects to router B which connects to router C, and that traffic has been rising on these links, straining their capabilities. The traditional approach would be to increase the link interface speeds, and in addition perhaps add processing power and buffer memory to intermediary router B. All expensive propositions.
With a holistic multilayer approach, we quickly see that in fact a significant amount of the traffic is in fact packets being routed from A to C, which are just transiting B. Further, we discover that by reshuffling resources we can use the optical network to engineer a new connection directly from A to C for this traffic, bypassing B. This uses relatively less-expensive optical resources instead of router resources.
By multiplying this simple example a thousand-fold to reflect the much more extensive and complex service demands of the real world, it is easy to see how network-wide resources can be optimised.
Another role of multilayer optimisation is policy alignment. All networks are governed by various high-level policies such as fibre fill load balancing, shortest routes, maximum number of node hops, transit delays, and shared risk groups that ensure that primary and backup paths do not share any common fibre throughout their lengths. In the current situation, it is difficult to track when these drift out of spec, and even once this is discovered fixes tend to be isolated patches. By using an SDN-based multilayer approach that is continuously scanning the overall network situation, it is much easier to catch problems early on and implement improvements to keep all network layers humming together as a fine-tuned machine.
An outstanding attribute that end-users expect from telecommunications services and the networks that support them is very high availability. At the highest level, we refer to five nines or 99.999 per cent availability, which is a downtime of only five minutes per year. Achieving high availability, however, comes at a very high cost.
In packet networks the usual standard is that every core network router must be connected to at least three other routers, so as to be able to withstand at least two separate link failures. In other words, if a link fails, the router software will initiate routing of that port’s traffic in another direction. Similarly in optical transport networks most links use some form of 1:N protection, where N is often 1, so that resources are sitting completely idle simply waiting to be activated in the event of a failure. Clearly this adds a huge expense compared to a world where no failures occur or where long service outages were tolerated.
This is further exacerbated by the current approach where protection and restoration schemes are localised to the packet and transport networks. This provides an opportunity for multilayer restoration approaches to dramatically reduce capital costs. In fact, various quantitative studies of multilayer optimisation and restoration indicate typical capex savings of more than 30 per cent from the status quo, with the bulk often coming from the restoration piece.
Multilayer restoration implements savings through two basic approaches. The first is outright elimination of excess ports on routers, by reducing for example the minimum number of ports on a core router from three to two ports. By homing these router ports onto an OTN switch or a reconfigurable optical add-drop multiplexer, in the event of a link failure, the SDN controller can immediately direct packet traffic through an alternate optical network path instead of relying on a pure router layer solution.
The second approach relies on migrating Layer 0 wavelength-switched optical network (WSON) and Layer 1 automatic switched optical network (ASON) restoration architectures from their current independent distributed control to integrated centralised SDN control. This provides a big picture view that is much better equipped to recognise and implement restoration schemes, both within the transport network, and working with the packet and other services layers.
The cloud is changing users’ expectations from services. They want a portal interface through which they can order services and have them turned up, turned down, or modified, in real time. SDN-based multilayer provisioning is a key for enabling this experience in the telecom world. Multilayer provisioning uses real-time multilayer path computation and software control to rapidly create cost-effective paths for new Ethernet, Fibre Channel, video, time-division multiplexed and other service connections, without human intervention. This increases customer responsiveness and reduces operations expenses.
Even more, multilayer provisioning enables new services and revenue streams, often based on variations of bandwidth on demand. One example is a dynamic data centre interconnect (DCI) service. While data centres already have fixed links among themselves, they often require immediate short-term, high-bandwidth connection for applications like cloud bursting, unplanned backups, or data and virtual machine migration. Data centres subscribing to a dynamic DCI service would have a basic connection to a service provider network. Using dynamic DCI, these data centres would be able to obtain any-to-any short-term, high-bandwidth connections among themselves, and essentially only pay for the bandwidth that they use and when they use it. This would supplement their fixed connections and provide them with redundancy in the event of failures.
Another service can be termed dynamic SD-WAN. Software defined WANs provide enterprises with optimised cost-performance WANs by combining non-deterministic but less expensive broadband networking with deterministic but more expensive MPLS VPNs. A limitation of SD-WANs today is these can only optimise existing broadband and MPLS services, and cannot adapt to short term needs or gracefully evolve as traffic conditions change. A dynamic SD-WAN service would be able to modify bandwidth of the underlying services dynamically, to provide an even more effective enterprise WAN experience.
Taking this a step further, when SDN-based multilayer provisioning is combined with another important modernisation initiative, network functions virtualisation, we are on a firm footing to achieve Network-as-a-Service. This is the cloud vision and holy grail of extending to end users an ability to dynamically ‘dial up’ all manner of network services, and have them available in minutes.
The current approach of managing packet and optical transport networks is based on the historical parallel evolution of these networks. Now that data traffic is dominant it is time to bring them together into an optimised multilayer framework, enabled through centralised SDN intelligence and control. Multilayer optimisation and multilayer restoration provides significant capital expenditure savings. Multilayer provisioning can create a cloud-like services experience. It speeds up service delivery, saves on operations expense through streamlining and automation, and above all enables new classes of bandwidth services and revenue streams.
Jonathan Homa is senior director, portfolio marketing, at ECI Telecom
- Multi-Layer Capacity Planning for IP-Optical Networks, Gerstel et al, IEEE Communications Magazine, January 2014
- Multi-layer Restoration - The Impact on the Optical Layer, Matthias Gunkel, 2014 Optical Society America
- Survivable IP/MPLS-Over-WSON Multilayer Network Optimization. M. Ruiz et al, O. Pedrola, L. Aug 2011/J. Opt. Commun. Netw.