Application-Specific Secure Gathering of Consumer Preferences and Feedback in Information-Centric Networks Reza Tourani, Satyajayant (Jay) Misra, Travis Mick Computer Science Department New Mexico State University New Mexico State University, NM
Outline v Introduction and Motivation v Problem Definition v Requirements and Preliminaries v Feedback Collection and Delivery Approaches v Conclusions and Future Work New Mexico State University, NM
Outline v Introduction and Motivation v Problem Definition v Requirements and Preliminaries v Feedback Collection and Delivery Approaches v Conclusions and Future Work New Mexico State University, NM
Client mining is widespread. New Mexico State University, NM
Benefits of client mining and recommender systems. Influence on 80% of hours streamed at Netflix (2016) Approximately 35% increase in Amazon revenue (2013) 50% of LinkedIn job applications and job views by members (2011) New Mexico State University, NM
Netflix Communication Flow Redirecting users Amazon Cloud CDN Routing Netflix Data Center DRM ( www.netflix.com ) Netflix Server Authentication Manifest File Periodic Updates CDNs New user registration User account billing Client New Mexico State University, NM
Client-Server Interaction in Hulu s.hulu.com (Hosted by Akamai) 3 CDNs www.hulu.com t.hulu.com (Hosted by Akamai) (Hosted by Huku) Client New Mexico State University, NM
Outline v Introduction and Motivation v Problem Definition v Requirements and Preliminaries v Feedback Collection and Delivery Approaches v Conclusions and Future Work New Mexico State University, NM
Request flow in ICN multi-level architecture. New Mexico State University, NM
Data flow in ICN multi-level architecture. New Mexico State University, NM
Pervasive caching eliminates contacting provider for popular content. Cache Hit! New Mexico State University, NM
How to track client without communication? Caching undermines gathering of access statistics. New Mexico State University, NM
Outline v Introduction and Motivation v Problem Definition v Requirements and Preliminaries v Feedback Collection and Delivery Approaches v Conclusions and Future Work New Mexico State University, NM
ICN requirements for successful client mining. Precise Statistics Preserving User Privacy Content Provider Independent Secure feedback collection New Mexico State University, NM
Content Categorization Static • Generated in advance • Generated in advance Content • Available publicly • Require access control • Cacheable • Cacheable Generated Beforehand Require Public Private Publicly Access Content Available Content Control Generated On-Demand • Generated by request • Generated by request • Require access control • Available publicly Dynamic • Non-Cacheable Content New Mexico State University, NM
Static-Public is the largest content category. Content Type in North America 34% 66% Static-Public Other New Mexico State University, NM
A bigger portion of mobile access traffic is encrypted in comparison to fixed access traffic. Fixed Access Traffic in Mobile Access Traffic in Fixed Access Traffic in North America 2015 North America 2016 North America 2016 29% 36% 37% 63% 64% 71% Encrypted Un-Encrypted Encrypted Un-Encrypted Encrypted Un-Encrypted Spotlight: Encrypted Internet traffic (https://www.sandvine.com/trends/global-internet-phenomena) New Mexico State University, NM
Outline v Introduction and Motivation v Problem Definition v Requirements and Preliminaries v Feedback Collection and Delivery Approaches v Conclusions and Future Work New Mexico State University, NM
Feedback Collection and Delivery Preference Tracking Mechanisms Manifest-Free Manifest-Based Collection by Manifest Collection by Collection by Manifest Intermediate from ISP's Clients ISP's Server from Provider routers Server New Mexico State University, NM
Drawbacks of Collection by Intermediate Routers Collection Event Per-Interest Per-Hit Drawbacks Drawbacks Redundant Statistics Coarse-level Statistics Coarse-level Statistics Computation Overhead Computation Overhead Lack of Client ID Lack of Client ID New Mexico State University, NM
Drawbacks of Collection by Clients Approaches Access Control Content Partitioning Enforcement Drawbacks Drawbacks Unknown Partition Size Suitable for Private Content Partition Publication Dependency on Online Server Dependency on Online Server Communication Overhead Communication Overhead New Mexico State University, NM
Collection by the ISP’s Designated Server Provider ISP's Server User • Offload • Stores Statistics • Request Decryption Key Decryption Key • Returns or Content or Content Requested Key or Partition Partition Content Benefits Drawbacks Reduced Latency ISP-Provider Interaction Cache Utilization Inaccurate Statistics Independent of Provider New Mexico State University, NM
Manifest-Based Approaches Manifest Delivery Provider ISP’s Server (Un-cacheable) (Cacheable) Drawbacks Drawbacks Extra Latency Single Point of Failure Provider’s Availability Network Bottleneck Un-cached Content New Mexico State University, NM
Outline v Introduction and Motivation v Problem Definition v Requirements and Preliminaries v Feedback Collection and Delivery Approaches v Conclusions and Future Work New Mexico State University, NM
How about evaluation? v Manifest-based approaches scale much better than the other schemes. v Communication overhead as means of evaluating efficacy of the approaches – Manifest based approaches scale better. v Manifest-based approaches introduce fixed amount of overhead per content and the amortized cost will be low. v There is a theoretical upper bound on the required communication overhead per content: Overhead = 𝐸𝑗𝑡𝑢𝑏𝑜𝑑𝑓 𝑑𝑚𝑗𝑓𝑜𝑢 − 𝑡𝑓𝑠𝑤𝑓𝑠 ✕ 𝑁𝑏𝑜𝑗𝑔𝑓𝑡𝑢𝑡𝑗𝑨𝑓 New Mexico State University, NM
Conclusions and Future Work v Direct interaction between client and provider with/without help of routers inaccurate and non- scalable. v A viable feedback collection mechanism should leverage caching. v Manifest-based feedback collection approaches are more scalable, especially if it involves infrastructure at the ISP. v Comprehensive evaluation of manifest based approaches (Provider vs. ISP server) and identify which approach in the other class comes closer. New Mexico State University, NM
Thank you! Email:misra@cs.nmsu.edu Research funded by the US National Science Foundation and the US Dept. of Defense. New Mexico State University, NM
Recommend
More recommend
