From TCP to Net Neutrality and Back Aleksandar

From TCP to Net Neutrality and Back Aleksandar Kuzmanovic EECS Department Northwestern University http: //networks. cs. northwestern. edu

Today’s Talk TCP congestion control Reusing CDNs’ measurements Net neutrality 2 A. Kuzmanovic From TCP to Net Neutrality and Back

TCP Congestion Control Question – Why do we care about TCP congestion control in the year 2007? Overwhelming opinion: – – TCP research is incremental Not relevant any more It is boring No high-impact breakthroughs are possible any more 3 A. Kuzmanovic From TCP to Net Neutrality and Back

Non-Incremental Advances are Possible “… throughput increases by more than 40% while the average web response time simultaneously decreases by nearly an order of magnitude. ” Server A. Kuzmanovic, “The Power of Explicit Congestion Notification, ” in ACM SIGCOMM 2005. A. Kuzmanovic, S. Floyd, and K. K. Ramakrishnan “Adding Explicit Congestion Notification (ECN) to TCP’s SYN/ACK Packets, ” IETF Draft, work in progress. 4 A. Kuzmanovic From TCP to Net Neutrality and Back

TCP Congestion Control • Slow-start phase • Double the sending. . . rate each round-trip. . . time • Reach high throughput. . . quickly 5 A. Kuzmanovic From TCP to Net Neutrality and Back

TCP Congestion Control • Additive Increase –. . . Multiplicative Decrease • Fairness among flows 6 A. Kuzmanovic From TCP to Net Neutrality and Back

TCP Congestion Control • Exponential • . backoff • System stability 7 A. Kuzmanovic From TCP to Net Neutrality and Back

Breakthroughs are Possible • Exponential backoff • fundamentally wrong! A. Mondal and A. Kuzmanovic, “Removing Exponential Backoff from TCP, ” work in progress. 8 A. Kuzmanovic From TCP to Net Neutrality and Back

Today’s Talk TCP congestion control Reusing CDNs’ measurements Net neutrality 9 A. Kuzmanovic From TCP to Net Neutrality and Back

Drafting Behind Akamai Fact – CDNs (e. g. , Akamai) perform extensive network and server measurements • Publish the results through DNS over short time scales Can overlay networks reuse measurements collected by production CDNs? – Significantly reduce the amount of measurements (a complementary service) – No new infrastructure need to be deployed – Inherit the robustness of DNS – Easy integration with existing systems 10 A. Kuzmanovic From TCP to Net Neutrality and Back

CDN-Driven One-Hop Source Routing D A 1 … … E 1 . . A 2 E 2 An En S DNS Server 11 A. Kuzmanovic From TCP to Net Neutrality and Back

Results Key findings – DNS redirections sufficiently small – Strong correlation to network conditions – 50% of nodes “discovered” by Akamai outperform direct paths Global Internet “weather-report” service for little to no cost About impact – Akamai never liked the idea… A-J. Su, D. Choffnes, A. Kuzmanovic, and F. Bustamante “Drafting Behind Akamai (Travelocity-Based Detouring), ” in ACM SIGCOMM 2006. 12 A. Kuzmanovic From TCP to Net Neutrality and Back

Relative Network Positioning via CDN Redirections Wide-area distributed network systems can benefit from network positioning systems – P 2 p data sharing, multi-cast, online games Relative network position is sufficient for most applications – Relative order is more important than absolute distance 13 A. Kuzmanovic From TCP to Net Neutrality and Back

Closest Node Selection Redirection frequency for node A to replica server y Redirection frequency vector for node A cos_sim(A, B) < cos_sim(A, C) => d. A, B > d. A, C 14 A. Kuzmanovic From TCP to Net Neutrality and Back

Rank Comparison 25% CRP Top 1 closely matches Meridian 20% CRP Top 5 outperforms Meridian 15 A. Kuzmanovic From TCP to Net Neutrality and Back

Latency Analysis 80% of CRP Top 5 Error < 50 ms 50% of CRP Top 5 Error < 25 ms 16 A. Kuzmanovic From TCP to Net Neutrality and Back

Clustering No common replica server between 2 clusters Node {D, E, F} redirect to common replica X 17 A. Kuzmanovic From TCP to Net Neutrality and Back

Central Leader Election Node B has the largest sum of cosine similarity 18 A. Kuzmanovic From TCP to Net Neutrality and Back

Today’s Talk TCP congestion control Reusing CDNs’ measurements Net Neutrality 19 A. Kuzmanovic From TCP to Net Neutrality and Back

Net Neutrality Pro net neutrality: – www. savetheinternet. com – http: //www. youtube. com/watch? v=v 4 c. Yuin. TGWA Anti net neutrality: – http: //www. handsoff. org/blog/ 20 A. Kuzmanovic From TCP to Net Neutrality and Back

What is this all about? Google, Skype Microsoft, Yahoo, etc. AT&T 21 A. Kuzmanovic From TCP to Net Neutrality and Back

Net Neutrality: Reality ISPs deny service to Vo. IP flows ISP 1 ISP 2 ISP 3 All networks with sufficient marketing power apply similar methods – The problem is moving deeper into the Internet core – The same is happening to p 2 p, gaming, and streaming services Not just about discrimination: – Entire policies may be tuned in a biased way 22 A. Kuzmanovic From TCP to Net Neutrality and Back

Networking Research Give peace a chance! – How can ISPs cooperate for their mutual benefits? – How can endpoints optimize their local behavior while maximizing network performance? – How can we provide incentives for cooperation in other networking domains? But what are we going to do if there will be no peace soon? 23 A. Kuzmanovic From TCP to Net Neutrality and Back

Our Approach Internet Audit: – A distributed system to enable network accountability: • What happened, where did it happen, and who is responsible? Challenges: – Non-repudiatiable identification of discriminating network elements – Detect unfair service favoring, e. g. , content provider/ISP alliances – Explore a range of threat models • from open Do. S attacks to using network policies in destructive ways 24 A. Kuzmanovic From TCP to Net Neutrality and Back

Preliminary Results Problem: – Detect precise locations where queuing happens • A key building block in detecting discriminating jitter-boxes S D – Useful to know in general: • Overlay design (route around such spots) • Advanced congestion control • Fault diagnosis 25 A. Kuzmanovic From TCP to Net Neutrality and Back

Coordinated Probing Probe f S s d D b f probe, b probe, s probe, d probe 4 -p probing: a symmetric path scenario 26 A. Kuzmanovic From TCP to Net Neutrality and Back

Coordinated Probing Probe f S Δd s d Δs Δb Δfs Δfd Δf D b Locating Tracing Congestion Points Status Half-path queuing delay 27 A. Kuzmanovic From TCP to Net Neutrality and Back

Methodology Highlights Coordinated probing – Send 4, 3, or 2 packets from two endpoints Quality of Measurability (Qo. M) – Able to deterministically detect its own inaccuracy Self-adaptivity – Switch among different probing schemes based on Qo. M and path properties 28 A. Kuzmanovic From TCP to Net Neutrality and Back

Evaluation 1 1 2 0. 37 s on/off 2 3 3 4 4 5 5 6 6 7 7 8 8 0. 71 s on/off 9 9 10 10 11 11 12 0. 53 s on/off 0. 29 s on/off 0. 63 s on/off After adding two more forward bottlenecks 29 A. Kuzmanovic From TCP to Net Neutrality and Back

Results Edge vs. core – Edge more frequently congested than the core: 4. 5 times on average Intra-AS vs. Inter-AS – Edge: Intra-AS > Inter-AS – Core: Intra-AS < Inter-AS Time domain – Edges: congestion events clustered in time – Core: congestion events dispersed in time Links vs. Paths – Links: 12% congested, 3% considerably – Paths: 20% considerably congested 30 A. Kuzmanovic From TCP to Net Neutrality and Back

Multiple Congested Points Probability to observe multiple congested points on an end-to-end path – Grows as a power function of interval length – Decays exponentially with the number of congested points 31 A. Kuzmanovic From TCP to Net Neutrality and Back

Summary Congestion control Reusing CDNs’ redirections Net neutrality and congestion measurements (Ericsson, Cisco) Other projects Online traffic classification and profiling (Narus) Monitoring p 2 p misconfigurations http: //networks. cs. northwestern. edu 32 A. Kuzmanovic From TCP to Net Neutrality and Back