Peer pressure

Brokered Private Peering could ease some of the internet's reliability problems by giving smaller ISPs a way of guaranteeing quality of service. Davey Winder reports.

Brokered Private Peering (BPP) could be among that rare breed of USity problems by giving smaller ISPs a way of guaranteeing quality of service. Davey Winder reports. network hype that actually changes the way the internet works. This fast-evolving peering model provides a method for medium-sized Internet Service Providers (ISPs) to exchange traffic reliably, without compromising either Quality of Service (QoS) or Service Level Agreements (SLAs).

A number of US players have become founding members of the BPP Group, including infrastructure companies like Williams Telecommunications, as well as smaller backbone providers Exodus Communications and Electric Lightwave. Ascend Communications has also signed up as a sponsor to help with technical assistance and hardware.

So how does peering work? In order for the internet to function properly, data packets must be able to reach their destination efficiently. Unless the start and end points are both within the same network backbone structure, at some point the data will need to cross to another backbone to complete its journey. Public exchange points where these backbones can interconnect, known as Network Access Points (NAPs) and Metropolitan Area Exchanges (MAEs), form the basis of public peering. These have struggled to keep up with the growth of the internet and, as such, public peering points have been responsible for many of the backbone bottlenecks we experience.

To overcome this, some of the major ISPs have entered into private peering agreements, where they connect their backbones directly to each other for an exchange of data.

Share and share alike

Public peering is where a backbone provider connects its network to a shared data exchange point at an MAE or NAP, with many peers being mapped to its network on each link. Private peering is where backbone providers of similar size and traffic flow enter into private data exchange agreements between their networks.

BPP is an internet peering exchange architecture using an ATM switch to interconnect three or more partners, providing a shared interconnect with dedicated bandwidth. Public peering cannot offer any QoS guarantees because it is open to all comers while private peering is more of an exclusive club available only to the biggest players with the biggest budgets. BPP brings the advantages of the private peering model to a wider audience of smaller backbone providers.

The current state of play sees a straight choice between private peering and public peering, the former being done by a privately negotiated agreement, usually between carriers of similar size with similar traffic flows over a dedicated loop; the latter by way of shared NAPs or ISP co-operatives such as LINX.

The BPP Group aims to bring the benefits of shared peering that NAPs offer, but with the same type of requirement for network availability and performance that is demanded by carriers in private peering agreements.

The plan is to have six ATM-based peering points initially in the US, with more to follow both domestically and internationally as the model is established.

The BPP Group proposal has fairly strict membership rules that require 99.9 per cent network reliability, zero per cent packet loss and zero per cent latency, together with even tighter controls over when ISPs have to add capacity or move to a proper privately peered (as opposed to brokered) arrangement. The new peering model promises to reduce the bandwidth burden of data congestion and so increase internet throughput. The hope, eventually, is for QoS and SLAs that can be implemented across several carriers' networks.

Next stop Europe

So will these guaranteed end-to-end links be the future model for internet traffic in Europe, or is this going to remain strictly a Stateside innovation?

Mark Purdom, marketing manager at Ascend Communications, is in no doubt that BPP has to be the new European peering model. "Although we do not have the internet infrastructure of the US at present, the buildout is rapid," he said. "The agreements will almost certainly be driven by the expansion of the US-based carriers into Europe and their need to sign such agreements with the existing European carriers. BPP is not all about speeds and feeds, but rather the exchange of traffic between carriers."

Neil McRae, operations manager at Colt Internet, agreed, but argues it's already happening. "We had an offer from one European service provider to do this and we currently peer with Deutsche Telekom privately."

However, not everyone agrees. Robin Duke-Woolley, consultant at independent telecoms consultancy Schema, paints a different picture. He pointed out that there are large numbers of smaller ISPs without national coverage in the US and therefore it is reasonable to expect that BPP should gain a sufficient following that would establish the infrastructure required over there.

The European situation is rather different, of course, with the telcos either having already, or being on the verge of having, substantial IP networks in their respective countries. Add to this the fact that there are typically five or six ISPs with significant network infrastructures in each country and it becomes obvious that for BPP to succeed in Europe it needs to provide a solution at the international rather than national level.

"It would mainly appeal to small, national ISPs that cannot provide end-to-end service within their own networks and do not have the resources to extend their networks," said Duke-Woolley. "These are less likely to succeed in the large corporate environment in any case. As presently defined, BPP is likely to remain a strictly US affair."

Many would argue that we don't need BPP in the UK. The big players are already wrapped up in their private peering arrangements, while the small to medium-sized ISPs either hang off these large backbone providers or join one of the available co-operatives like LINX.

But Purdom sees problems with both models. "They do not scale to the internet," he said. "As an example, within BPP agreements there will be clauses looking at QoS guarantees, whereby multiple services are run over a single network ATM infrastructure and when handed off to another carrier, the QoS parameters are maintained."

Duke-Woolley leans more toward QoS, explaining that public peering at LINX and other exchange points in Europe concentrate on efficient connectivity and do not seek to provide end-to-end QoS between ISPs. "Private peering has developed between larger ISPs to extend their overall reach of fast internet access and managed network services. As in the US, smaller ISPs are unlikely to benefit from this unless they happen to use these ISPs as their backbone networks - for which they must pay," he said.

LINX chairman Keith Mitchell challenges Purdom's view that LINX does not scale to the internet. "Private peering does not scale, N players require (N2+1) divided by two links, which is expensive. Private peering has the potential for abuse by large players that may insist on large settlement payments from smaller players to have access to their customer base. We have a number of measures at LINX to ensure peering fair play."

One apparent concern is the future of the small ISP in the face of commercial peering, no matter how it is wrapped up - BPP, private peering agreement or otherwise. Can these small players survive in the face of such competition from corporate giants?

There seems little doubt that their position will be weakened by the increased use of private peering arrangements where they will have little to offer the big boys whose bandwidth they want to share. The cost of moving from a public peering model to a private one is prohibitive to these smaller players and effectively presents an insurmountable barrier.

The warning shots were fired early last year when UUNet in the US looked at changing the way its peering relationships worked. Although the changes were never implemented, the plan to introduce requirements of national, diversely routed and dedicated 45Mbps or higher backbones, connecting to UUNet at speeds in excess of 45Mbps in at least four geographically dispersed sites, would have a significant impact on the smaller ISPs.

Survival of the biggest

Purdom believes that, in the face of these changing peering models, the only future for smaller ISPs is to be acquired by the larger players or die trying. "Bear in mind that a diversification of service providers will happen," he said. "Look at Williams Communications in the US; its model is to be the carriers' carrier, to build out a backbone infrastructure which they use to resell pure bandwidth with QoS guarantees to downstream carriers. Such a model is very appropriate in Europe."

This would mean that ISPs need to start thinking more about both marketing and market position, and start thinking about it now. As consumer internet services no longer seem to be a good enough business model for ISPs to survive on, the more market-focused players will be looking to partner bandwidth providers so as to be able to offer business services as a competitive differentiator. "The US guys already see this model," Purdom said, "and are clamouring to get over here to sell their services."

It would seem to be the case that BPP calls for some over-engineering of the network in order to ensure reliability, and that in itself may be yet another barrier to its success.

McRae doesn't agree. "Private peering requires no over-engineering. It does require additional infrastructure, but this can often be simple fibre connections at appropriate hubbing locations."

Mitchell also doesn't think there is necessarily any connection between the issues of performance and peering models. "In my experience, internet growth remains such that demand usually exceeds supply and the engineering realities take precedence over whatever is on any paper agreement."

Internet Exchange Points (IXP) such as LINX have a proven track record of keeping costs down for the ISPs concerned and these savings are passed on to the end user. At the end of the day, BPP is all about providing a higher quality of service to the end user - and QoS always comes at an increased cost.

In the US, SLAs are fast becoming huge business and BPP will enable the provision of more meaningful SLAs for the internet. The model is to bring the reliability of data networks up to the same level demanded of voice networks. As Ascend's Purdom commented: "All vendors that want a place in this market must be able to offer equipment with the highest possible up-time. Those that can't will not survive." Let's just hope that the cost of the internet coming of age isn't too high.

PEERING MODELS: EUROPE VERSUS THE US

LINX chairman Keith Mitchell gives his thoughts on the future of peering models in the US and Europe.

LINX, the London Internet Exchange, handles more than 98 per cent of UK internet traffic and, as such, is the most efficient path to over 15 per cent of all internet routes. Its 45 members, which come from seven countries, include Demon Internet, Easynet, GX Networks and UUNet.

"The situation in Europe is rather different from the US. In the US, most of the exchange points/NAPs are run by telcos. In Europe, they tend to be run either by the public sector (off the back of the national academic network, although decreasingly so) or as ISP co-operatives like LINX.

"I believe the US model has led to under-developed Internet Exchange Points (IXPs) as the telco owners are more interested in maximising carrier business into the IXP rather than investing and managing the IXP fabric. Worse, the MAEs and NAPs are now mainly owned by those telcos that own the most dominant ISPs in the US.

"I believe this to be a conflict of interest at best and the main reason for under-development of the US exchanges at worst.

"Because of this, US IXPs are, in general, over-subscribed and poorly managed, giving this model an undeserved bad name.

"Another major driving force for private peering in the US is large telcos flexing their market muscle and attempting to undermine and remove the open playing field for competition that exists at IXPs. It is notable that the one major IXP in the US which is not run by telcos, the Digital PAIX, does not suffer from the scaling and management problems of the NAPs and MAEs.

"Also, to date, no US IXPs have deployed Gigabit Ethernet technology to help with their scaling problems, something we will be doing at the LINX in the next few months.

"The other important difference between Europe and the US is the way the single European telecoms market is being implemented.

"To sum up, I believe there is room for both models, but private interconnect is less likely to be dominant in Europe."