FIND PI Meeting, April Contract-Switching: Value Flows in Inter-Domain Routing Murat Yuksel University of Nevada – Reno Reno, NV Aparna Gupta, Koushik Kar, Shiv Kalyanaraman Rensselaer Polytechnic Institute Troy, NY Project Website Or google Contract Switching
FIND PI Meeting, April Implied Challenges Motivation Current architectural problems: Users cannot express value choices at sufficient granularity – only at access level Providers do not have economic knobs to manage risks involved in QoS investments inter-ISP relationships flexibility in time: forward/option pricing flexibility in space: user-defined inter- domain routes capability to provide e2e higher quality services money-back guarantees, risk/cost sharing
FIND PI Meeting, April Inter-domain struggles… When crossing domains, all bets are off.. End-to-end QoS or reliability requires assurance of single-domain performance, i.e., contracts efficient concatenation of single-domain contracts Inter-domain routing needs to be aware of economic semantics contract routing + risk management We address translation of these struggles to architectural problems
FIND PI Meeting, April Contract-switching: A paradigm shift… Circuit-switching Packet-switching Contract-switching ISP A ISP C ISP B e2e circuits ISP A ISP C ISP B routable datagrams ISP A ISP C ISP B contracts overlaid on routable datagrams
FIND PI Meeting, April Contract Link An ISP is abstracted as a set of contract links Contract link: an advertisable contract between peering/edge points with flexibility of advertising different prices for edge-to-edge (g2g) intra-domain paths capability of managing value flows at a finer granularity than point-to-anywhere deals
FIND PI Meeting, April How to achieve e2e QoS? Contract Routing – spatial composition of e2e contracts Compose e2e inter-domain contract paths over available contract links satisfying the QoS requirements Calculate the contract paths by shortest-path algos with metrics customized w.r.t. contract QoS metrics Two ways: link-state contract routing at macro time-scales path-vector contract routing at micro time-scales Monitor and verify that each ISP involved in an e2e contract path is doing the job Punish the ISPs not doing their job, e.g. as a money-back guarantee to the others involved in the e2e contract path
FIND PI Meeting, April Link-State Contract Routing: Macro-level, proactive User X ISP A ISP C ISP B 1 Owner ISP LinkQoSTermOffered After Price ($/term) A1-210Mb/s2hrs1hr$10 A1-340Mb/s5hrs15mins$80 B2-4100Mb/s3hrs2hrs$110 C3-520Mb/s1hr30mins$8 C4-560Mb/s1day2hrs$250 4 Most cost- efficient route Max QoS route Global Internet 2008
FIND PI Meeting, April Path-Vector Contract Routing: Micro-level, on-demand, reactive User initiates… User X wants to know if it can reach 5 with 10-30Mb/s for mins in a $10 budget User X ISP A ISP C ISP B 14 [5, A-B, 1-2-4, Mb/s, 20-30mins, $4] [5, A, 1-2, 15-30Mb/s, 15-30mins, $8] [5, 10-30Mb/s, 15-45mins, $10] [5, A, 1-3, 5-10Mb/s, 15-20mins, $7] Paths to 5 are found and ISP C sends replies to the user with two specific contract- path-vectors. path request [A-B-C, , 20Mb/s, 30mins] [A-C, 1-3-5, 10Mb/s, 15mins] Paths to 5 are found and ISP C sends replies to the user with two specific contract- path-vectors. reply Global Internet 2008
FIND PI Meeting, April Path-Vector Contract Routing: Micro-level, on-demand, reactive Provider initiates… ISP C wants to advertise availability of a short-term contract link User X ISP A ISP C ISP B 14 [C, 5-4, 30Mb/s, 45mins, $9] [C-B, 5-4-2, 20Mb/s, 45mins, $6+$5] [C-B-A, , 20Mb/s, 30mins, $7.3+$3] [C, 5-3, 10Mb/s, 30mins, $5] [C-A, 5-3-1, 5Mb/s, 15mins, $1.25+$1.2] path announcement path announcement path announcement Global Internet 2008
FIND PI Meeting, April Temporal Extensions of Single- domain QoS Contracts Bailout Forwards: on advertisable spot contracts with flexibility of advertising different forward prices for g2g intra-domain paths Forwards with provision for bailout conditioned on network status Time
FIND PI Meeting, April 2009 Bailout Forward Contract Multiple g2g Contracts Exposed to the additional effects of other traffic flows as a result of overlapping links with other g2g overlay paths How do we model this interaction?
FIND PI Meeting, April 2009 Intensity of Overlap Link Capacity 1.5 Gbps Utilization = 10% Link Capacity 0.3 Gbps Utilization = 50% Flow 1 50 mbps Flow mbps
FIND PI Meeting, April 2009 BFC Performance Can an ISP survive by applying BFC approach? How frequent does BFC bailout? BFC Robustness against increasing demand? (demand) decreasing available bandwidth? (supply) major link failures ? How efficient is BFC pricing? Revenue Losses
FIND PI Meeting, April 2009 Simulation Results: Dynamic Demand Path Forward Prices E[S T ] Probability {S T >F} Probability {A T <Th} Threshold = 15 th percentile Results for five sample g2g contracts Financial guess – Forward breakdown prob Technical guess - Bailout probability IEEE IWQoS 2008
FIND PI Meeting, April 2009 Simulation Results: Dynamic demand Mean fraction 16.4 % EXODUS – 372 g2g paths IEEE IWQoS 2008
FIND PI Meeting, April 2009 Network Analysis: Robustness against failures Mean fraction 27% EXODUS – 372 g2g paths IEEE IWQoS 2008
FIND PI Meeting, April 2009 Simulation Results: Effect of Simplified Pricing ABOVENET – histogram of 7 days revenue Node 5 Node 1 pt-anywhere pricing g2g pricing simplified g2g pricing
FIND PI Meeting, April 2009 Simulation Results: Effect of Forward Pricing ABOVENET – 40% of available capacity is contracted as forward Node 1 pt-anywhere pricing simplified g2g pricing simplified forward g2g pricing
FIND PI Meeting, April Future Work and Questions Protocol implementation and simulation of CR Balance between LSCR and PVCR ISP collaboration and competition pricing the inter-domain risk game theoretic analysis contract verification common punishment and rewarding laws Goal: more economics in inter-domain routing
FIND PI Meeting, April 2009 Thank you! THE END