10.Pando P4 P Laird Popkin


Published on

Presentation on P2P, ISP's and P4P from the European Peering Forum 2007. Note that the simulations (AT&T and Abilene) were performed using publicly available BGP routing data, not ISP-provided network maps.

Published in: Business, Technology
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

10.Pando P4 P Laird Popkin

  1. 1. ® Can P2P and ISP’s Work Together? Laird Popkin, CTO, Pando Networks November 2007 Pando Confidential
  2. 2. Overview • Pando & P2P – Business – Technology – What are ISP’s telling us? – How can we work with ISP’s? • P4P – Who is in the P4P Working Group? – The goals – Results so far – Next steps 2 Confidential
  3. 3. The Opportunity of P2P The Internet is the media delivery “Within five years, all platform of the future media will be delivered New technologies are needed to across the Internet.” scale the Internet for higher quality media delivery - Steve Ballmer, CEO Microsoft P2P networks present a D5 Conference, June 2007 disruptive market opportunity 3 Confidential
  4. 4. Maturing P2P Market Nascent P2P Market (before 2007) Commercial P2P Market (2007+) “Content Delivery” mechanism P2P becomes part of content “File Sharing” application delivery infrastructure Rogue technology Content owners prefer to buy Stand alone P2P applications integrated P2P + CDN solution Pando.com, free consumer app. Major content and CDN players select P2P technology partners 4 Confidential
  5. 5. Pando is a Proven Platform Total Installs >15M Daily Media Deliveries >500K 30 Day Active Nodes 30 Day Online Nodes 6 Confidential
  6. 6. Swarm Efficiency Managed P2P: Managed P2P: • Control Servers Control Servers – Secure, Coordinate, Secure, Coordinate, Control, ReportReport Control, CDN CDN Servers • Servers – Reliable Reliable • P2P Clients P2P Clients – Efficient Efficient 15 Confidential
  7. 7. How is Pando different from “generic” P2P? Managed Peer to Peer Infrastructure Secure, Controlled Central Server Efficient & Scalable Distributed Network Dynamic & Targeted Monetization Platform 8 Confidential
  8. 8. Pando is a Market Leader Pando Unique Value Proposition • CDN Peer Assist • Pando and CDN’s pay for transit to provide reliability, performance • P2P streaming • P2P boost for progressive download/playback of FLV, QT, WMV, etc. • Managed Network • Reporting, control, DMCA takedowns, etc. • Integrates into applications and web services (APIs) • Proven scale and network efficiency • More “edge” than any “Edge Network” • Partnering with CDNs and selling directly • Pricing and QOS disruptive to large and growing market 7 Confidential
  9. 9. CDN PeerAssist (Marketing Diagram) Performance and Efficiency Optimized Throughout Content Demand Curve Content Demand Large, popular files Hi Resolution content (HD) RSS media subscriptions Game/software downloads “CDN Peering” P2P + HTTP P2P HTTP 9 Confidential
  10. 10. CDN PeerAssist (Real Data) Network Efficiency By Swarm Size 100.00% 100.00% 94.00% 76.25% % of data 57.00% delivered by 52.50% peer cloud vs. CDN. 28.75% 14.00% 5.00% 1000+ 100 - 999 10 - 99 1-9 # of nodes in swarm 10 Confidential
  11. 11. “Pando may represent the future of media distribution.”
  12. 12. Pando’s Technical Differentiators Pando Technology: • Cross-platform – Core is portable C code. Supported on Windows, Mac and Linux. • Native binary – efficient, embeddable. Platform-native GUI’s. • Widely used – Well tested, proven scalability • Web embedded through standard protocols – support any browser, any OS, any media player. • Resilient p2p protocol uses best available channel: – Full P2P through HTTP tunneling through proxies, with auto-detection, – UDP and TCP hole punching through firewalls, etc. • User are in control – can uninstall, set limits, etc. Result = 75% long-term retention rate. • Partners are in control – Control all presentation, branding (Pando Media Booster). • ISP-friendly P2P – Cache server integrations, P4P WG. 14 Confidential
  13. 13. Pando Client Architecture Clients: • Pando stand-alone • Pando Media Booster • Pando Command Line Pando integrates into: • Web Sites • Applications • Devices • Content Management Pando is portable: • Windows • Mac OS X • Linux 17 Confidential
  14. 14. Network Protocols Layered Protocol • Control: – SOAP over SSL • Tracker – TCP or UDP • Data Transfer – TCP w/ STUNT – UDP w/ STUN – HTTP, Proxy 18 Confidential
  15. 15. What is Pando hearing from ISP’s? Good: – P2P is a key reason that many customers get broadband – Customers love Pando • Some ISP’s are distributing “white label” Pando as a value-add to their customers Bad: – P2P consumes all available resources – P2P is costing us too much to support 20 Confidential
  16. 16. Pando and ISP’s What is Pando doing to work with ISP’s? • Pando pays for transit (for Pando-hosted content) • Unique protocol signature (Pando doesn’t hide) • Working with caching server vendors: – Caching servers have up to 80% “hit rate”, provide great performance – But require ISP capital investment, so can’t be the only answer • Make software smarter about ISP infrastructure – Attempts to reverse engineer structure are weak: • AS mapping • /24 mapping • Ping times • Traceroutes – So let’s get accurate knowledge from the ISP’s! • Thus P4P. 21 Confidential
  17. 17. P4P P4P: Partnership between ISP’s and P2P Networks Laird Popkin, Co-Chair, P4P Working Group 22 Confidential
  18. 18. P4P Working Group P4P Working Group: Co-Chaired by Pando and Verizon, based on research from Yale, hosted by Distributed Computing Industry Association. Observers Core Group Abacast AT&T Pando Networks Cablevision BitTorrent RawFlow CacheLogic Cisco Systems Telefonica Group Cox Communications Grid Networks VeriSign Comcast MPAA Joost Verizon NBC Universal LimeWire Washington Oversi University Manatt PeerApp Yale University Time Warner Cable Turner Broadcasting 23 Confidential
  19. 19. Roadmap • Motivation • P4P framework • Design rationale • System architecture • Computing peering suggestions • Evaluations • Ongoing work 24 Confidential
  20. 20. P2P : The Significant Bandwidth Consumer • Traffic – Up to 60-70% of Internet traffic is contributed by P2P applications [CacheLogic] – Random peering causes traffic spread across POPs and domains • Problems – Increased network resource usage (e.g., using bandwidth of more links) – Increased network operational costs – Degraded performance of other applications 25 Confidential
  21. 21. Bandwidth Battle between ISPs and P2P ISP’s Address P2P P2P Countermeasures Upgrade Network Use random ports Infrastructure Encrypt traffic Deploy P2P Caching etc. Devices Terminate User Connectivity Rate Limit P2P Traffic etc. The battle results in a lose-lose situation 26 Confidential
  22. 22. Where is the Fundamental Problem? • Traditional ISP feedback/control to application traffic: – Routing – Rate control through congestion feedback (packet drops) • These are ineffective for P2P – Due to highly dynamic, scattered traffic pattern caused by dynamic, unguided (network-oblivious) peer selection • Need a mechanism for ISPs to communicate with P2P about network structure and policies 27 Confidential
  23. 23. Objective • Design a framework to enable better ISP and P2P coordination – ISPs and P2P jointly decide P2P peer selection • Guided P2P connections should yield benefits: – Improve throughput to P2P users – Allow ISP’s to manage link utilization • Reduce congestion for capped links • Reduce link costs for non-capped links • Push traffic from (expensive/limited) external links to internal links 28 Confidential
  24. 24. Network Awareness redefines P2P Network Aware P2P will reduce costs, improve performance Traditional CDN Network Aware P2P Platform Most Limited Internet Transit Regional Network Most Available Edge Network More Viewers = More Viewers = Worse performance Better performance Lower cost Higher cost 29 Confidential
  25. 25. P4P Framework – Goals • Performance improvement for both ISPs and P2P • Scalability – Support a large number of P2P users and networks in dynamic settings • Privacy preservation • Flexibility: apply to many P2P architectures – Application-specific requirements – Tracker-based and trackerless P2P systems – “Gossip” among peers • Ease of implementation (“low hanging fruit”) • Open standard: any ISP, P2P can easily implement it 30 Confidential
  26. 26. P4P in Tracker-based P2P (e.g. BitTorrent) • Use BitTorrent in a single pTracker iTracker ISP as an example 2 – pTracker runs P2P system – iTracker makes suggestions for peering relationships • Information flow: 3 – 1. peer queries pTracker – 2. pTracker asks iTracker for 4 1 guidance (occasionally) – 3. iTracker returns high-level peering suggestions – 4. pTracker selects and returns a set of active peers, according to the ISP A suggestions peer iTracker can be run by trusted third parties, or P2P networks, or ISP’s. 31 Confidential
  27. 27. Compute Suggested Peering Relationships • Formulate as a joint optimization problem – ISP’s objective: minimize maximum link utilization – P2P’s objective: maximize throughput 33 Confidential
  28. 28. Results Evaluated P4P approach via simulations and experiments: • Simulations – Discrete-event simulation • a module for modeling BitTorrent protocol • a module for modeling underlying network topology and data transfer dynamics using TCP rate equation – Network topology: PoP-level Abilene and AT&T topologies – Network routing: OSPF routing • PlanetLab experiments – 53 Internet2 nodes on PlanetLab – iTracker for Abilene network – Use OSPF routing to re-construct traffic load on Abilene links 34
  29. 29. Abilene Simulation Results: Good for P2P Networks Evaluation – Abilene Simulation P4P yields over 2x speed increase! 35 Confidential
  30. 30. Abilene Simulation Results: Good for ISP’s Evaluation – Abilene Simulation P4P yields over 2x drop in external link utilization! (and for ½ the time!) 36 Confidential
  31. 31. AT&T Simulation Results Confirm Confirmed on AT&T model: • P4P resulted in: – 1.6x faster completion – 3x lower link utilization • Native P2P can result in some peers experiencing very long download completion time • P4P can result in much lower variance in link utilization 37 37 Confidential
  32. 32. PlanetLab Implementation Confirms Link Utilization BT BT+P4P Swarm Size Implemented on PlanetLab: • Implemented P4P in BT Tracker, etc. • Run on 53 PlanetLab nodes for 900 seconds • Liveswarms is BT-based video streaming • P4P resulted in: over 2x lower link utilization 37 37 Confidential
  33. 33. Interested? Interested in P4P? P4P WG is free to join. Email marty@dcia.info or laird@pando.com Monthly phone calls Working Group mission: Evaluate the P4P design through large-scale experiments Formalize and promote adoption of P4P protocols Serve as a forum for ISP’s and P2P networks Questions? 38 Confidential