PlusNet's Traffic Management Roadmap

Industry background

1. The growth of broadband
2. The future for broadband
3. Practical implications
4. Consumer usage
5. Business usage
6. Asymmetrical data transfer
7. The social networking effect
8. The video effect
9. The P2P effect
10. Contention
11. How does this all add up?
12. Where is this headed?

PlusNet's traffic management solution

1. General background
2. So what have we done so far?
3. The need for intelligent networks - PlusNet's approach
4. Conclusion
   
   
   

Industry background

In the first section of this article we explain how and why the explosion of Broadband from an early adopter proposition to a mass market one has caused problems for service providers generally.

1. The growth of broadband
The advent of universally available broadband has enabled the growth of the Internet to continue almost unabated up to this point. Broadband access in the UK is no longer a specialist product used only by IT professionals and technology enthusiasts. As a mass market commodity, broadband is as important to many people as owning a mobile phone or car.

As broadband take-up accelerated, and demand began to outstrip supply, the wholesale charging models for broadband have also changed. The flat rate charging system, which ignored usage was replaced by capacity based pricing structures. This was advantageous in that it lowered the cost of entry for new consumers and stimulated market adoption of higher speed products. For a time, ISPs tried to absorb the extra costs, which made this change relatively invisible to customers. However it has become clear that with increasing speeds and very diverse usage requirements, a one size fits all product is unfair on many, while allowing others to significantly drive up costs for an ISP. These “all you can eat” models need to be replaced by fairer and more understandable ways to pay for broadband access.

Most people are familiar with the traditional sources of high demand placed on Broadband networks. High bandwidth video consumption, initially via P2P (Filesharing) has grown immensely in recent years, and the trend has been accelerated more recently by applications like streaming radio and TV, as well as websites like YouTube and MySpace. Increasing 'average usage' means ISPs now need to reach to the very depths of their technical expertise, and their wallets, to scale existing networks to support these demands. All broadband ISPs have had to adopt measures in one form or another to control their capacity and ensure they are not paying more for overall customer usage than customers are paying them.

The diagram below shows the emergence of Internet protocols which have impacted on capacity usage over the last 10 years. (The trend has continued in the same direction since 2004):

The situation many ISPs now routinely face is their broadband networks running at or near full capacity for large parts of the day. Whenever that is the case it opens the door to many potential problems. These include exposure to infrastructure failures or a sudden surge in demand. Even with anticipated future reductions in wholesale costs we believe that pressure relating to demand will never be fully offset by underlying cost reductions.

2. The future for broadband
There is no shortage of discussion about the applications with mass market appeal which are due on-stream in the next few months and years. All manner of technical innovations are currently in the early stages of life, from networked social computer games through streaming interactive High Definition TV to remote access and the next generation of web applications arriving via Web 2.0. It is clear from experiences in the US and Asia that while not all of these applications will take off, many millions of people will adopt at least some of these new technologies, and will come to rely on them over the next few years.

Everything we have listed has the need for a high quality broadband connection in common. At any one time this connection will need to seamlessly support these applications running in parallel, potentially in multiple rooms of a house. Of course, the drivers of demand for broadband so far are also still gaining momentum alongside these new applications. How ISPs design their products and platforms to cope with this growth is one of the biggest questions being asked of the ISP industry globally.

In terms of broadband in the UK, the next generation telecommunications infrastructure, 21CN, is now being deployed by BT. Within a few years most people will have access to this new broadband network, with BT having committed to providing at least 50% of broadband connections over 21CN by the end 2007. 21CN has been designed around higher bandwidth applications like streaming video and Voice over IP, where the original broadband networks were designed for low bandwidth applications such as text emails and web surfing. However, if the growth of broadband so far is anything to go by, it seems that demand is going to continue to outstrip supply regardless of how much capacity can be provisioned by ISPs.

21CN is about more than simply delivering support for more demanding applications. BT is transforming its entire infrastructure into a pure IP-based network. As well as an increase in efficiency, this will allow customers to get instant broadband access wherever they are, using a range of devices. With BT pressing ahead on initiatives like Wireless City and (later) Wimax, wireless broadband access will become increasingly common. In the future it will be possible for people to use their own broadband login details over any of these different fixed or wireless networks. A customer’s roaming login will enable more than a simple connection too – Consumers will be able to access and control their entire network profile wherever they are and however they are connected.

3. Practical implications
One of the primary goals for any broadband ISP is to provide its customer base with reliable, always-on access to the Internet, so they can consume real-time applications any time of the day or night and be able to download large files without impacting real-time performance traffic for them, or any other users of the platform. That aim presents a multitude of challenges which the ISP must overcome. Most notably, the ISP must be able to determine what type of traffic is being consumed across its network in real-time so it can provide an appropriate level of service for it.

The first difficultly therefore becomes the identification of traffic. For an example of how complicated this can be, web based usage can incorporate lots of different types of data:

• HTTP on Port 80 for web pages
• Streaming on Port 80 for YouTube or similar video clips.
• Streaming on Port 80 for radio services such as BBC, Pandora.com or last.fm, either via web pages or hardware devices such as the Squeezebox.
• Gaming on Port 80 for services such as Second Life or World of Warcraft.
• Downloading on Port 80 for zipped USENET files.
• Download on Port 80 for Microsoft patches & updates.
• Downloading on Port 80 for large size file downloads such as Linux ISOs.
• Browsing on Port 80 for real time business applications such as salesforce.com.
• Use of software like Google Earth professional and other 'thin' clients.
  

From the list above, not all are required in real-time. Web pages, streaming services and business applications are interactive and as such require delivering as soon as possible. However, file downloads and patch updates do not require instant delivery and can be spread over a longer period of time before completing.

There are additional factors which also impact networks and how ISPs deliver traffic across them, as described below.

4. Consumer usage
The peak network demand relating to real-time traffic is driven by consumers, and is therefore highest during normal peak hours; i.e. after 3:30pm when school finishes and again after 6pm when other family members start to use the PC or other Internet based devices.

We are also seeing a shift in how people perceive the Internet. When people purchase devices like the Slim Devices Squeezebox, and stream radio stations this way, it’s easy to forget that you are consuming Internet traffic. As a result, people tend to leave these devices running (often turning the volume down rather than switching them off when not in use!). A radio stream running in the background at 128kbps for a typical 8 hour period is 460Mb per day or over 10Gb per month (based on an average 20 day working month). When that same customer starts to consume other types of traffic, such as Microsoft updates, and then also downloads a couple of albums from iTunes, the numbers continue to increase.

5. Business usage
Business users tend to consume traffic in a similar way to consumers for the majority of web based traffic. However, Business needs are shifting towards more web based applications like salesforce.com (an online CRM tool) as well as running bespoke applications over virtual private networks (VPNs). Here the speed of interaction and reliability is key. We are also seeing big growth in the usage of VoIP and video conferencing for routine business interaction, all of which needs to be catered for seamlessly.

6. Asymmetrical data transfer
The problem is set to become even more acute when we look at some of the newer devices becoming available, such as the Slingbox from Sling Media. This device is asymmetric, in that it actually moves the data in the opposite direction to most traffic. The Slingbox takes content from within the home network and pushes it out via the Internet so it can be consumed remotely.

7. The social networking effect
Newer devices, such as Nintendo’s Wii integrate gaming and community into the same environment. With services like the Mii avatar channel, the more people that join the community, the more data needs to be sent between devices to keep each "person" in synch. We expect this to be a common theme as more socially aware devices and features come onto the market.

8. The video effect
Receiving data from a web page, predominately made up of text and graphics, has been a feature of the Internet since the beginning. Recently, the web has adopted a number of enhancements, such as advertisements driven by Flash streaming or new service such as Flickr. This is predominately images and not text, and especially with sites like YouTube, where all of the content is streamed, video creates a situation where each web page is now orders of magnitude larger in size than traditional text and image pages.

This is further exacerbated when we consider the advent of new devices such as IP enabled set top boxes where complete TV programs and schedules are downloaded or streamed via the Internet to the TV.

9. The P2P effect
P2P (Filesharing applications like BitTorrent and Kazaa) by their very nature are "dirty" protocols. This means they will consume as much of the available bandwidth as is available - give them an inch and they will try to take a yard! This has driven up total bandwidth consumption worldwide, as shown in the earlier diagram.  Furthermore, the saturating effect of these applications will detrimentally impact the quality of all other traffic being consumed on the network at the same time. P2P is predominately a consumer orientated service, currently used to download music, videos or software (both legitimately and illegitimately).

10. Contention
Businesses want to consume their data typically during the normal business hours of 8am and 8pm. Consumers are more complicated. Broadly speaking, consumer use can be categorised into groups like 'family', 'single' and 'gaming'. Family traffic tends to be consumed between 3:30pm and 7:30pm. Single type traffic tends to be consumed between 7pm and 10pm. Gaming content typically gets consumed between 7pm and Midnight.

This presents an incremental 'kicker' situation for usage during the day, where the network is a blend of all of these types of users and usage profiles. As a result, these users become contended against the network for all of their types of usage. I.e. the business customer using his online application at salesforce.com at 7pm is using the network at the same time as the family who are watching YouTube video streaming in one room and making a VoIP call to family in Australasia in the other, all at the same time as the gamer is playing their Xbox 360 with friends across the globe.

11. How does this all add up?
Based on the above, we can see that certain times of the day are more popular than others for usage.  i.e. between 7pm and 10pm is a busy time for many types of users consuming many different types of traffic. This then becomes the most contended part of the day for non-interactive traffic, and as such the value of a packet of data at this time of day is considerably higher than that provided at 3am.  If an ISP were to charge on a "Supply & Demand" basis, people would be pay something like £4 per GB during the peak times of 6pm and 10pm, and a nominal amount per GB for the same packets at 3am.

12. Where is this headed?
Sooner or later all ISPs are going to need to provide an environment where there is not just prioritisation of data, but also that data is tagged for desirability. This will complicate things for ISPs, but will ultimately make the service much more valuable and rewarding for customers. By integrating traffic management, broadband firewalls, content control and parental control facilities into one platform it is possible to offer a fully customisable and configurable services which suit every type of customer.

To take this to the next level, not only might we need to know that the data being transferred is an image which should be delivered immediately, we also need to understand what type of image it is and whether it’s desirable for that image to actually be delivered at all.

For example, a family could have decided that the Internet can only be used on the PC in the living room between 3:30pm and Midnight, but between 3;30pm and 7:30pm it is a restricted service where content only from a white listed set of sites can be delivered and all others are blocked. As such, content of a certain classification can never get delivered to that PC during 3:30pm and 7:30pm, but as soon as the 7:30pm watershed has been reached all types of image content can be delivered.

These are the sort of changes that will put customers in control, and provided that the interface for customer use is well designed, the complexity behind the scenes can be made transparent to customers. It’s no trivial task to achieve this, but it is something we are already developing, and which we feel will become an important aspect for consumers when they consider their broadband provision. It's important to add that as an ISP, we have no desire to police what content is consumed - we want to enable our customers to have that control themselves.

PlusNet's traffic management solution

The second section of this article explains how we are planning to evolve our broadband traffic management strategy, in order to support the growing demands for broadband in the future.

1. General background
Since it became apparent that ISPs would be unable to continue to support ever increasing usage demands, in a world that was changing from fixed costs to capacity based charging, we have been working on finding solutions to enable us to offer the best possible value for money and performance to our customer base.

2. So what have we done so far?
Towards the end of summer 2004 when the new wholesale charging model was used, it became clear that expenditure on capacity was accelerating much faster than growth in revenue could support. That change had enabled the roll-out of affordable higher speed services, initially in the form of 2MB, but it also presented us with the immediate challenge of bringing the heaviest usage on our platform under control. We did this using the fairly limited tools available to us at that time, and we put in place a blunt configuration which segregated capacity for this group of users.

Our next step was to implement principles of product group differentiation, meaning that customers were given a product experience relative to the bandwidth contribution in the subscription they paid. We then developed a process which tied in the amount of capacity each customer paid for directly with their product type and usage. You can see more about this in The PlusNet Broadband 'Blueprint'. A combination of monitoring heavy users, variable rate limits and protocol prioritisation were used to achieve this. This was only developed initially as a semi-automated system which managed our network and controlled the sharing of available capacity fairly for each customer. By April 2006, we had automated these management mechanisms, and we have continued to develop that automation ever since.

We spent the latter half of 2006 working on ensuring that we were using the available technology in the most effective way possible, and we are now arriving at the next stage in the evolution of our strategy. Our approach so far has been to adopt relatively generic network rules, applied to different product groups and protocols. Although comparatively fair, the problems with this approach have become apparent to some customers as average product usage has increased and people’s requirements have become more wide-ranging.

We have invested in equipment which prioritises packet delivery and ensures that the most time sensitive data is fast-tracked even when the network is very busy. Our traffic prioritisation system provide a more graceful experience for our customers than they would receive from an ISP who didn’t adopt such techniques when handling network contention, whether that was caused by an outage or happening for another reason. Again though, so far this technology has only been delivered to customers on a per product group basis, with different products receiving different priority for their traffic and generic decisions being made about what constitutes the most important, time sensitive, traffic.

While our traffic management techniques have been effective at allowing us to keep costs under control, and there have been benefits delivered to consumers in terms of performance of time sensitive application, it’s clear that many consumers still regard traffic management with trepidation. To understand why PlusNet has put so much time and effort into developing the technology behind traffic management, you have to think back four years and then consider where we are today and how much things have changed.

3. The need for intelligent networks – PlusNet's approach
As described above, running a broadband network in a financially stable manner at the same time as delivering a good experience is no small challenge.

We have already built a solution which allows us to provide a good level of service to the majority of our customers for most protocols currently identified. We now need to provide a good level of service for the majority of customers when using every protocol.

An intelligent network is needed to achieve this. That network needs to know what’s happening and who is using what, not yesterday, but right now (or at least within the last 15 minutes). This will allow it to adjust and make decisions on what needs to changes in order provide a good level of service to the majority of users currently connected.

We are designing a solution where the “pulse” of the network can be taken in real-time (15 mins or less), and the system can react to the current situation through a loop back mechanism called a "heartbeat". This could be described technically as the first step towards a "stochastic" solution, which works in a similar way to a runners body when they use a treadmill. As the runner changes their speed or incline, their body automatically reacts to the changing conditions (increasing the size of the lungs, heart rate, etc) to ensure the fastest distribution of blood around the body. A "stochastic" network does the same thing, adapting to changes in demand and making intelligent decisions in real-time. In network terms, this means the platform can automatically flex so that it can accommodate the maximum number of current users for the majority of protocols being consumed.

For example:
If 10% of the customers are consuming P2P traffic without impacting the protocols of the other 90% of customers, then the network will allow that to continue.  However, if at any stage the 10% starts to induce a negative experience, then P2P will be squeezed appropriately to enable a better service for the 90% of customers who are not using it. If 15 minutes later the 90% of customers leave the network (e.g. There's an England world cup game on TV), then the network will change so that 10% of customers can utilise all of the available resources until a time when they are required again.

By making these small and immediate changes to the platform every 15 minutes or so, we can accommodate a wider range of services and provide a better experience for the majority of users. Over time we would like to bring that 15 minutes down to 2 or 3 minutes, but there is a cost associated with this level of granularity which is not currently viable (as technology becomes cheaper then this will become possible).

4. Conclusion
Having proportioned our network fairly, and completed a significant amount of the development work on the systems already, we can now focus on delivering more of the benefits that traffic management can offer to our customers. We are in a position where we can talk about traffic management controlled largely by the customer within the product, as opposed to people having to choose a specific products that best suits their individual needs. The aim is to make it possible for customers to easily select and control their own traffic based on their own personal requirements at any particular time.

This IEC technical white paper provides one view of current thinking in the industry, but we are already looking ahead of this. Traffic control can provide customers with the tools they need to manage their own bandwidth experience in a way that suits them. There can never be enough fairness in one size fits all profiles for customers who have very differing requirements.

During the next 12 months, PlusNet should be ready to deploy its next generation of traffic control systems, which will provide much greater flexibility for customers. As well as our move towards a "stochastic" broadband network, we are also working on the parental control and security functionality described in this document. That work includes building the easy to use interfaces which are so important given that we don't expect customers to need to understand the details of protocols or prioritisation in order to select how they want their broadband experience to be delivered.

This is a work in progress, development continues and we roll-out enhancements to our network management systems regularly. As a result there will invariably be times when this causes temporary impact to some customers. While our technology greatly enhances our ability to dynamically manage the network and assure the experience for any customer at any time, we know that there will always be imperfections. We are keen that customers should always be aware of the status of our network and of any factors that may be causing a difference in performance. The Managing our Network Under Abnormal Load page contains useful information about this.

Related links

Read the PlusNet Broadband Blueprint
Find out about our plans for the Future of Traffic Management

Rate this page

Has this page helped you solve your problem? Your feedback helps us to improve the help we provide.