© 2008 Nokia V1-Filename.ppt / YYYY-MM-DD / Initials 1 Dealing with P2P Traffic in an Operator Network: State-of-the-Art Hannes Tschofenig Marcin Matuszewski P2P Infrastructure Workshop, May 28, 2008

© 2008 Nokia V1-Filename.ppt / YYYY-MM-DD / Initials 2 History Bob Briscoe’s proposal around “flow rate fairness” submitted to the IETF TSVWG lead to some discussions. Bob proposed solutions to deal with congestion in an e2e fashion with reECN. We then thought it would be appropriate to investigate what mechanisms are available in theory today and what mechanisms are actually being used today. We have spoken to around 20+ ISPs (mostly European ISPs) and have interacted with the Finish regulator Klaus Nieminen to determine the state of the art. We would like to thank them for their feedback. We understand that this is not a scientific result since the data is somewhat difficult to obtain. The Finish regulator was able to run a survey whereby the ISPs had “incentive” to provide an honest response. A similar survey throughout Europe is, however, neither available nor planned. We believe it is important to understand what the “Best Current Practice” is before jumping into technical solutions.

© 2008 Nokia V1-Filename.ppt / YYYY-MM-DD / Initials 3 Situation Profit margins are low Situation different throughout different parts of the world (e.g., US vs. Europe) depending on the competitor situation Subscriber acquisition cost much higher than revenue of several months Cost differs based on available infrastructure, density of population and degree of competition (marketing costs) Difficulty to quickly change contracts with customers Example: Difficulty to add “fair usage policy” to an existing contract Heavy hitters are a perceived problem. The increase of traffic due to P2P file sharing may not have been expected and hence statistics did not work out for some operators: “Assuming that nobody would take advantage of a ‘flat-rate’ might have been wrong.” Some papers hint to 5% of the users generating 75% of traffic, 20% of the users generating 90% of traffic. Please note that measurements may vary significantly.

© 2008 Nokia V1-Filename.ppt / YYYY-MM-DD / Initials 4 Where does the cost come from? Fixed Networks, such as DSL 1.Wire to the subscriber’s home 2.DSLAM-to-BRAS backhaul 3.Transit Transit creates the highest costs. A lot of P2P traffic is not local. Even other traffic may not be local either but depends on the country. Example: Finland – 60% of the traffic is transit. How large costs are for (1)+(2) depends on regulation, type of business agreement, etc. For the incumbent (1)+(2) less problematic since this is his own network. Exact prices for (1)-(3) are difficult to obtain since they largely depend on non-public business agreements. In some cases the prices for (1) and (2) are regulated. Cellular Networks 1.Radio Access Network (site costs) 2.Base Station-to-GGSN backhaul (typically a mixture of own and leased lines) 3.Transit Cost of (1) and (2) very high; transit (in comparison) less of a problem. Note that a single subscriber in a cellular network can impact other subscribers to a certain extend due to the scheduling that is being applied. In DSL networks the subscribers traffic is often routed towards the BRAS and not directly extended between users (for security reasons and due to ATM usage).

© 2008 Nokia V1-Filename.ppt / YYYY-MM-DD / Initials 5 What todo if traffic increases? Available Tools Provide more bandwidth Volume based accounting Mainly in mobile networks Customers prefer flat-rate pricing Often: “flat-rate” (called unlimited data rate) scheme combined with an upper bound. Shaping all traffic if upper limit is reached. Charging for excessive traffic (mobile networks) Note: Whatever the ISP does it typically has to be described in the contract. (e.g., fair usage policy). When enough competition is in place then this causes fewer problems with the regulator. Standard Solution Done in mobile networks Rarely done Done in mobile networks

© 2008 Nokia V1-Filename.ppt / YYYY-MM-DD / Initials 6 Tools cont. Discontinue contract Block Subscriber Typically only used in context of DDoS mitigation Deep packet inspection to deal with specific type of traffic Example action: Shaping of BitTorrent Giving higher QoS to new flows (for a short period of time) Limiting subscriber flows Banning servers from residential access Grouping of subscriber classes and let them “fight” for resources among themselves (within their QoS class) Maintaining Super Peers or/and Content Cashes Legal aspects are likely to be challenging. Very rarely done. Done by some ISPs Done by 1 ISP Very rarely done Written in contracts but not enforced. Nobody told us that they deploy super peers --- they would be crazy todo so. Very rarely done. Done by 1 ISP

© 2008 Nokia V1-Filename.ppt / YYYY-MM-DD / Initials 7 Finland Case* 1 st Study: March 2004 (40 ISPs provided answers) ISPs were not blocking traffic at application level or applying any caching or query redirection solutions. ½ of the ISPs banned P2P usage by banning the subscribers to maintain servers as part of the contract. Was, however, not enforced. 20 % of the ISPs were policing P2P traffic at least for some of their connections. ( by prioritizing the traffic) Many ISPs indicated interest to consider usage of various tools to deal with heavy hitters. 2 nd Study: December 2007 (44 ISPs provided answer) Quite a few operators have any sort of mechanism in place. 1 operator boosts priority of new connections temporarily. Regulator does not seem to see a need to take actions based on the current situation. *: Data has been provided by Finish regulator, Klaus Nieminen

© 2008 Nokia V1-Filename.ppt / YYYY-MM-DD / Initials 8 Conclusions Heavy hitters do exist. Many ISPs consider them as a problem. A fair number of tools are available already. However, many operators claim that this is a business / marketing problem rather than a technical problem. Finland example shows that ISPs use the “dump more bandwidth into the network” approach. Do we need more tools? No tool is without cost. So, how expensive is your proposal going to be? What is the “user experience” going to be? What about legal aspects (contracts, copyright), marketing (bad press), etc.?