210 likes | 371 Views
P4P: Provider Portal for Applications. Haiyong Xie, Y. Richard Yang Arvind Krishnamurthy, Yanbin Liu, Avi Silberschatz SIGCOMM ’08 Hoon-gyu Choi hgchoi@mmlab.snu.ac.kr 2008.10.06. Contents. Introduction P4P Evaluation Summary. P2P: Bandwidth usage.
E N D
P4P: Provider Portal for Applications Haiyong Xie, Y. Richard Yang Arvind Krishnamurthy, Yanbin Liu, Avi Silberschatz SIGCOMM ’08 Hoon-gyu Choi hgchoi@mmlab.snu.ac.kr 2008.10.06
Contents • Introduction • P4P • Evaluation • Summary MMLAB
P2P: Bandwidth usage • Up to 70% of Internet traffic is contributed by P2P applications • However, the emerging P2P applications expose significant new challenges to Internet traffic control Ipoque Study 2007 Internet Protocol Breakdown 1993 - 2006 MMLAB
P2P Problem: Network inefficiency • Network-oblivious P2P applications may not be network efficient • Verizon • Average P2P bit traverses 1000 miles • Average P2P bit traverses 5.5 metro-hops • Karagiannis et al., BitTorrent on a University network (2005) • 50%-90% of existing local pieces in active users are downloaded externally MMLAB
Attempts to address P2P Problems • ISP Approaches • Increase capacity • Pricing • Rate limit/terminate P2P traffic • Deploy P2P caching devices • P2P Approaches • Locality aware P2P MMLAB
P2P Problem: Inefficient interactions • ISP optimizer interacts poorly with adaptive P2P • ISP • Traffic engineering to change routing to shift traffic away from highly utilized links • Adaptive P2P • Adapt their traffic to changes in the network • Resulting in potential oscillations in traffic patterns and sub-optimal routing decisions • Traditional Internet architectural feedback to applications is limited • Emerging P2P applications can have tremendous flexibility in shaping how data is communicated • The network needs to provide more information and feedback to most effectively utilize this flexibility for improving network efficiency MMLAB
P4P • Design a framework to enable better providers and applications cooperation • P4P: provider portal for (P2P) applications • A provider can be • A traditional ISP (e.g., AT&T, Verizon) • A content distribution provider (e.g., Akamai) • A caching provider (e.g., PeerApp) • Open standard • Any ISP, provider, application can easily implement it MMLAB
P4P Objectives • ISP perspective • Guide applications to achieve more efficient network usage, e.g., • Avoid undesirable (expensive/limited capacity) links to more desirable (inexpensive/available capacity) links • Resource providers (e.g., caching, CDN, ISP) perspective • Provide applications with on-demand resources/quality • P2P perspective • Better performance for users • Decreased incentive for ISPs to “manage” applications MMLAB
P4P Architecture • P4P Potential entities • iTracker: individual network providers • Peer: P2P clients • appTracker: P2P • Each network provider maintains an iTracker MMLAB
P4P iTracker • iTracker • A portal for each network resource provider • Allows P4P to divide traffic control responsibilities between applications and network providers • Makes P4P incrementally deployable and extensible • iTracker of a provider can be identified in various ways • e.g., through DNS query • iTracker can be run by trusted third parties • Examples of iTracker interfaces • Policy interface • Allows applications to obtain the usage polices of a network • Provider capabilities interface • Allows peers to request network providers’ capabilities • P4P-distance (Virtual cost) interface • Allows others to query costs and distances between peers MMLAB
P4P distance • P4P-distance reflect the network’s status and preferences regarding application traffic • P4P-distance should be • Simple, intuitive, … • An ISP can assign P4P-distance in a wide variety of ways • OSPF weights and BGP preferences • Considering financial costs or approaching congestion • ETC. MMLAB
The P4P-distance Interface: two views • Networks’ view seen by an iTracker • The higher the price, the more “cost” to the ISP if an application uses the link • Reflects both network status and policy, e.g., • Higher prices on links with highest util. or higher than a threshold • OSPF weights • Applications’ view seen by applications • Applications adjust traffic patterns to place less load on more expensive P2P node pairs • Both ISP and Application can use this interface in a variety of ways MMLAB
appTracker iTracker 2 3 4 1 peer An example of P4P information flows • Information flow 1. peer queries appTracker 2/3. appTracker asks iTracker for virtual cost (occasionally) 4. appTracker selects and returns a set of active peers, according to both application requirements and iTracker information MMLAB
Evaluation Methodology • Network Topologies • Internet experiments on Abilene and ISP-B • Simulations on PoP(Point of Presence)-level topologies of Abilene and major tier-1 ISPs • Applications • BitTorrent, Liveswarms (streaming) and Pando (commercial) • Performance Metrics • Completion time • P2P bandwidth-distance product (BDP) • P2P traffic on top of the most utilized link • Charging volume MMLAB
Evaluation for Intra-domain • Simulation using Bit Torrent on ISP-A • P4P achieves rate between latency-based localized and native • The utilization of P4P is less than one-half of localized, which achieves lower than native Completion time Bottleneck link utilization MMLAB
Evaluation for Inter-domain • Experiments using BitTorrent on Abilene • P4P achieves similar application performance with localized; but P4P has a shorter tail. • For the charging volume of the second link: native is 4x of P4P; delay-localized is 2x of P4P Completion time Charging volumes MMLAB
Evaluation – Field Tests • Field Tests on ISP-B against Native (Pando) • P4P achieves approximately 5 times in unit BDP • ISP Perspective • P4P improves average completion time by 23%. • P2P Perspective Average Unit BDP Completion time MMLAB
Summary • P4P: provider portal for (P2P) applications • Simple and flexible framework • Explicit cooperation between P2P and network providers • P4P can be a promising approach to improve both application performance and provider efficiency MMLAB
References • Open P4P • http://www.openp4p.net/ • Yale P4P • http://cs-www.cs.yale.edu/homes/yong/p4p.html • P4P Working group, • http://www.dcia.info/activities/p4pwg/membership.html MMLAB
Discussions • What are the incentives for P2P to participate in P4P? • Better network efficiency • P2P by playing nice could avoid being blocked by ISPs • P4P leaves much flexibility for P2P • Benefits the overall society • Why cannot P2P achieves the benefits of P4P by itself? • Probing the network to reverse engineer information such as topology and status is difficult • Cost and policy is difficult to reverse engineer MMLAB
Discussions • Does P4P violate network neutrality? • ISPs and P2P applications mutually agree to participate in P4P • How can it be feasible for P4P to orchestrate all these networks? • iTracker interfaces are light-weight and do not handle per-client application • Do the locality-aware P4P techniques reduce robustness? • P4P does not limit the mechanisms for improving robustness • If iTrackers are down, P2P applications can still make default application decisions MMLAB