Many people now think of cloud as probably the ultimate in virtualisation. But of course there are still many elements of ICT that are only beginning to feel the impact of virtualisation, much less be actually virtualised. Networking is the stand-out example. The preferred term now is ‘Software Defined Networking’ or SDN rather than a V acronym, not least because virtual networks and VPNs will still exist in the new software defined world that is certainly coming right across all ICT.
Why do we need to virtualise the network anyway, when fibre to the desk/home is practicable and 4G speeds will soon be in Gigabits? The Open Networking Foundation, just over two years old and already the leading body like other such independent foundations in ICT, states bluntly that today’s "traditional network architecture is ill-suited to meet the requirements of enterprises, carriers and end-users." Even as the hardware engineering continues to evolve and advance, our ability to make best use of the infrastructure is constrained by inadequate standards and protocols. This has become more important-and potentially critical for the Internet-because of the explosion of content, especially video and colour images, mobile data, the imminent ‘internet of things’ and of course Big Data generally. Genomics, astronomy and astrophysics, meteorology and many other fields of science can stress networks with data volumes never seen when the traditional tiered switch architectures were developed.
Traffic and control planes
The concept underpinning SDN is simple: if the data or traffic carrying plane and the control plane are de-coupled the formerly static network can become intelligent, responsive, programmable and centrally controlled. All of that can then be automated according to logical rules in response to traffic patterns or types or emergencies. Network devices, notably switches and routers, have typically involved some built-in intelligence but from multiple manufacturers and different and usually proprietary internal code. Network engineers now almost notoriously complain that a virtual machine can be spun up in a few clicks and minutes but it can then take hours or days of work to correctly configure and test the interfaces with the network.
OpenFlow is the ONF master protocol set for SDN, adopted from standards developed in Stanford and Berkeley and embraced at quite an early stage several years ago by Google and later Facebook. It is already widely adopted in the sense that if you order a network switch or router today the product delivered will almost certainly have OpenFlow compatibility built in or will provide for an upgrade. On the other hand, the network device manufacturers are going along with SDN if not reluctantly then with understandable caution. There is a clear and present danger to their Ethernet/IP environment. But they do recognise the inevitability of SDN because of the clear next generation functionality it will bring to everything from WANs to data centres to carriers. Not to mention us users, individually mere consumers, but collectively the market drivers.
Architecture dependent
"We have been dependent on much the same networking architecture for 45 years plus, some of it still showing its derivation from the original Arpanet," says Dan Pitt, executive director of the ONF and a veteran of networking technology in IBM, HP and Nortel. "Originally, the goal of the Internet was survivability, which meant that all of the intelligence had to be present in every node. But there is only so much intelligence that can live in every device. SDN offers the answer in principle for the demands today and tomorrow. It is a total change in ICT engineering culture-and a welcome democratisation-so that the network itself becomes programmable and adaptable."
Like our interviewees from the ICT sector, Pitt accepts that SDN will take another two, three or more years to become the default for all networks and the soft/hardware to run and manage them. "There examples today, more visible and open than Google, where SDN is already the norm and demonstrating the benefits in coping with the range of demands that can be put on a network."
He points to a large California university with over 120,000 users from faculty to students to security and maintenance staff that is 95% BYOD and highly decentralised. It has almost every known type of data traffic and volumes from email and text to full-motion HD video and Big Data processing. That means it faces more ICT complexity and especially network performance issues than most large or even global enterprises. SDN is proving a most successful tool in managing all of this and delivering high service quality and satisfaction levels across the campus, Pitt says.
Complex and error prone
Managing our networks today is complex, time-consuming, error-prone and costly in all terms. With, Dan Pitt explains, we can have an ultimate vision of entirely flexible and dynamic network management controlled by simple user interfaces. "They will be symbolic of the underlying infrastructure but with no special skills required. We can foresee policies and business rules written in plain language, for example, and even language-independent with symbols and graphics to allow control of everything in the network that is not already automated according to previously established policies."
Cisco is of obvious importance in any evolution of networking, as a tech leader and also as the vendor with probably the highest volume of installed systems and especially hardware. It is certainly on board the SDN movement conceptually and a member of the ONF. The Cisco Open Network Environment is described as a "strategy for network programmability" and offers choices in protocols and industry standards. This may not quite be the ONF gospel but the important point is that at this early adoption stage the industry leading network systems vendor is part of the broad consensus.
"We have been talking to our clients for over a year about SDN," says Ian Foddering, CTO of Cisco UK and Ireland. "Currently we see the ONF model as perhaps the classic and simple separation of the control and data planes. But we believe is that in the real market world a hybrid SDN solution is what most organisations will find most suitable and practical for some years yet. They will want to leverage what they already have but with new centralised control." Cisco is committed to open standards, he insists, pointing to the OpenDaylight consortium launched just last month. This initiative is under the wing of the Linux Foundation and includes most of the members of the ONF, with Cisco joined by Citrix, IBM, HP, Juniper Networks, VMware and others. ONF itself is not part of the consortium, although it seems its OpenFlow protocols will be adopted.
Maturing
Realistically, it will be at least three years before SDN reaches any serious scale in the market, Foddering says. Data centres and cloud service providers will lead the way, he believes, and some large global enterprises. "SDN will assist in creating massively scalable data centres and a greater level of automation, with both economic and efficiency benefits. We did a recent survey in Ireland and the UK that suggested there is actually an almost surprising level of clarity about what SDN has to offer. Organisations understand what it is and where it’s going and the potential benefits. It is to the questions about how to get there that we are all still developing the answers."
Simplicity of use and management is the key, he says, with the network complexity entirely hidden at the back end. "In some ways SDN should really be called ‘user defined networking’ which will drive more agility and responsiveness into networking management. We see it also as exposing network intelligence to the applications, again to add to the orchestration and automation. Deeper network intelligence can be pulled into a new class of analytical applications that can promote more sophisticated network policies and support business logic that drives the network," says Foddering. "This ultimately makes the network more valuable and can support more innovative and revenue-generating services."
Stuck in the past
VMware’s chief technologist Joe Baguley takes a blunt view: "Networking has stayed stuck in the past while we have made great forward strides in server virtualisation and cloud computing. Provisioning networking and security for a new VM or an application today is still a largely manual effort involving a human, keyboard and CLI." The key to automation, as with the other forms of virtualisation in ICT, is to abstract the software from the hardware. The vision of a software defined data centre is central to the next generation of ICT and at this stage networking is the drag factor in slowing development of almost everything else in our infrastructure.
"The overall trend of ICT development is clear," Baguley says. "We are moving towards a software defined and controlled environment, logical and ultimately flexible, running on essentially commodity hardware. Managing thousands of servers in an organisation has to become programmatic and move to policy-driven automation to deliver efficiency and flexibility, whether in scaling or in response to changing traffic needs."
All of that will progress the relationships between in-house computing and private and public cloud, he adds. Organisations will establish more clearly over time what is core and what is not in terms of data, applications and everything else. SDN as well as all other forms of virtualisation will enable the free and seamless movement of workloads according to the policies of the organisation and current needs. "Very stringent SLAs and security will certainly apply," says Baguley, "and in fact in a virtual and software-defined ICT world some service providers will undoubtedly begin to differentiate themselves according to their physical assets. The point is that we will have a full spectrum from public cloud down to handle any workload according to the policies and business rules of the user."
Hype and hindsight
We have clearly seen over recent decades that ICT is actually never mature and certainly never static, says Damian Saunders, director of the cloud platforms group in Citrix. "We can look back at all of the new technologies that caused waves of hype in their time and see with the clarity of hindsight that the successful ones have all been based on good science. Virtualisation has certainly been a good example, abstracting the software and control from the hardware on which it runs. Now we are as an industry belatedly bringing that concept to the network with a great deal of relevant expertise and technology already proven."
Saunders is another who believes ‘user defined networking’ might be a more meaningful term in many ways because SDN offers so much more than virtualisation and will bring policy driven flexibility to the ways in which we manage enterprise computing from now on. "It will enable, for example, much more elastic software licensing terms and turn the whole infrastructure stack into a programmable platform for users and information. We need to design more complexity into the data centre and the intelligence to present simplicity to users in the next generation of ICT."
Elastic capabilities
Elasticity is a key concept in this new context, says Saunders. "Traditionally, we had to forecast and plan for hardware capacity in advance, often for the probably lifetime of the asset. Of course resources could be added, but in general we were always buying a lot of capacity that we actually knew we were unlikely to use. ‘Bursting’ was an almost impossible notion. That was the old model. Now we provide capability on demand, roll it back if necessary and follow the normal ebb and flow of business and usage-not to mention revenue. Service provision and software licensing then can also follow the user-defined needs and on better terms. That is all genuinely transformative."
HP is very much still one of the world’s great multi-product, multi-service ICT giants and is involved with both the ONF and the Open DayLight consortium. Bruno Hareng is HP’s EMEA manager for SDN and explains that SDN is an inevitable response to the new challenges being faced by all kinds of networking by developments such as Big Data and the exponential demands of mobile consumer and business devices. "Legacy networks are at breaking point faced with the range of applications, traffic types and volumes today," he says. "We need to integrate servers, storage and networking into a common platform, centrally and intelligently controlled to recognise and serve the needs of all of the different-and changing-applications and workloads we need. Rigid architecture cannot do that."
Different strokes
There are a number of different approaches to SDN emerging in the market, Hareng acknowledges, but says that HP is adopting the OpenFlow definition and logic of uncoupling the control and data planes in the network architecture. "That is fairly generic and allows interoperability across ICT and telecommunications. Within the organisation, other approaches may be employed to achieve the same SDN characteristics and results in the infrastructure, notably application awareness and user-defined flexibility. Like so many other elements of ICT, real life is likely to generate mixed or hybrid solutions and we certainly still have to manage existing network devices."
Hareng points to the example of Universal Communications, rapidly growing as a key element in collaboration and imposing real-time demands on the network in tears of both voice and video. "Even ordinary users now understand something of the difficulty in managing network traffic to ensure priority and QoS for UC. But they also appreciate the value when it works well. SDN offers the possibility of doing the same thing for applications, enabling the network to respond dynamically to the demands of each application. Already, for example, SDN has been proven to dramatically improve the performance of Hadoop."
Home field
Here in Ireland we now have the infrastructure and bandwidth nationally to really exploit the potential of SDN according to Paul Phelan, CTO of networking specialists Data Edge. "We had to spend a few generations of technology learning our way, but SDN now sets us up for an era in which the network has brains. Traditionally it has had limited intelligence in discrete components. Put it this way: configuring a big router was-and indeed still is-one of the most dangerous jobs in ICT. There are enormous possibilities of error and misconfiguring and an entry error of one digit can bring down an entire network."
Quite rightly, in Phelan’s opinion, applications are now taking primacy across ICT and that is a natural and proper evolution. "In a sense we can ask why the hell this did not happen before now, but in fact there has been a kind of Darwinian evolution in ICT in recent years. We have developed robust and proven technology and the losers have gone and even the largely proprietary elements have largely lost ground. It is really fairly recently that virtualisation and related technology has been made robust enough to tackle the complexity of the network and especially its control layers."
SDN technology is already in place and well tested even if it is not yet mainstream in the market. "Google has been operating on SDN since 2010, the standards are there with OpenFlow and the performance and device testing-an area we are involved with-is also in place. SDN then offers almost limitless flexibility and scalability. Burst and surges? I can see service providers and carriers offering bandwidth on demand-by the girth by the minute. Software control delivers both performance and economy and competition will drive both."
Subscribers 0
Fans 0
Followers 0
Followers