190 likes | 281 Views
Hard Architectural Challenges and Initial Approaches. Arun Venkataramani Univ. Massachusetts Amherst arun@cs.umass.edu. Architecture: Design Goals. Host + network mobility No global root of trust Intentional data receipt Byzantine robustness Content addressability Evolvable network .
E N D
Hard Architectural Challenges and Initial Approaches ArunVenkataramani Univ. Massachusetts Amherst arun@cs.umass.edu
Architecture: Design Goals • Host + network mobility • No global root of trust • Intentional data receipt • Byzantine robustness • Content addressability • Evolvable network None of the above goals G1-G6 are not met by today’sInternet.
Location Service: Scalability to billions of mobiles • Function: Resolve Host [NA1, NA2,…] • Scale: 10B devices, 100 networks/day 10M/sec Locate(Host) Data Name servers Host[NA:NA1] Host[NA:NA2] NA2 NA1 NA
Location Service: Scalability to billions of mobiles • Function: Resolve Host [NA1, NA2,…] • Scale: 10B devices, 100 networks/day 10M/sec • Metrics: • Query/Update delay (<50ms) • Response staleness (<500ms) • Load balance • Fault tolerance Name servers
Location Service: Scalability to billions of mobiles • Function: Resolve Host [NA1, NA2,…] • Scale: 10B devices, 100 networks/day 10M/sec • Design issues: • In-situ routing deflection (?) • Structured local scope IDs (?) • Authoritative name servers (?) • Network anycast to name servers Name servers API
Location Service for Popular Content • Function: Resolve Content [NA1:HA1, NA2:HA2,…] • Design issues: • Ensuring proximate content retrieval • Leveraging traffic enggflexibility • Storage-aware routing using opportunistic caching/retrieval Name servers API
DT Routing: Robustness to Diversity • Goal: Protocol stack robust to diverse, changing network conditions Connectivity • Approach: • Hop-by-hop transport with in-network storage when needed • Gracefully degrades with challenging network conditions • Uncertainty-driven routing algorithms unifying storage & transmission capabilities of the network
E2E Routing: Enabling Path Diversity • Goal: Path diversity for • Optimizing performance, cost, power, security in multi-technologymobile connectivity • Approach: Multipath routing mechanisms • Multi-homing support • Detour routers • Randomized path “unchoking” • Coordinated path rate controller Management Plane ISP1 ISP2 path1, path2, path3 req_paths(…) ISP3
E2E Routing: Design Challenges R2 R3 R1 R5 R4 R6 • Store and update loose source routes, eg (NA:NA1:NA2) in a scalable manner • Supporting multiple interfaces • Each interface has an address. • Need to handle cases when interfaces change (e.g., the 3G unavailable). Bob Alice
Architecture: Design Goals • Host + network mobility • No global root of trust • Intentional data receipt • Byzantine robustness • Content addressability • Evolvable network None of the above goals G1-G6 are not met by today’sInternet.
Security: Hijack/Spoof tolerance • Alice wants to talk to Bob Name certification + location service Bob’s human readable name 2. Bob’s ID & address Routing Service 3. Send Alice’s handshake data signed by Alice’s ID To Bob’s address Alice Bob 6. Verify using Bob’s self-certifying ID 5. Send Bob’s handshake data signed by Bob’s ID To Alice’s address
Security: Decentralizing Trust in Naming • Goal: No single root of trust in name certification • Approach: • Multiple name service providers (NSP) • Many-to-many mapping from namespaces to NSPs • Quorum-based certification Name service 1 Name service 2 Name service 3
Security: No single point of subversion Goal: Scalably tolerating Byzantine faults for naming, routing Approach: • Naming: • BFT throughput scalability wrt faults and number of servers • Geographic scaling support • BFT within and across name service providers • Routing • Securable protocol design, eg, consensus interdomain routing, by identifying safety and liveness properties • Multipath + management plane for data plane security • Integration with naming, management plane, and intradomain routing
Security: Intentional receipt for DDos Congestion policing feedback (2) (1) Ra Rb Hd Hs congested mon mon mon nop (3) (4) Policing • Goal: Scalable fair resource allocationwithintentional data receipt • Approach: • Packets carry unspoofable congestion policing feedback • Congested routers use pair-wise keys for congestion policing feedback that receivers use as capability tokens • Access routers police senders’ traffic to guarantee per-sender fairness without per-sender queues
Privacy Challenges Goal: Quantifiable privacy Privacy Mechanisms Lookup Service NA:HA NA:HA Allow Home Agent Redirection Approach:Identifying privacy concerns: • Host identifiers allows linking traffic to specific devices • Self-certifying addresses reduce plausible deniability • Lookup service could enable geo-tracking by third parties Allow HA Pseudonyms Home Agent AS 1 AS 2 Hospital HA swaps pseudonym