100 likes | 437 Views
Data Oriented Network Architecture (DONA). Andrey Ermolinskiy Mohit Chawla CS 262 A Project Poster December 14. DONA Overview. DONA explores a clean-slate approach to Internet Architecture Main idea: data-centric routing:
E N D
Data Oriented Network Architecture (DONA) Andrey Ermolinskiy Mohit Chawla CS 262 A Project Poster December 14
DONA Overview • DONA explores a clean-slate approach to Internet Architecture • Main idea: data-centric routing: • Client requests a piece of data by its name rather than the owner’s address • Names are flat, self-certifying Data REGs FINDs • Dissemination Handlers (DH) are DONA “routers” • Combine forwarding, name resolution, and data caching functions
DONA Overview (cont.) • DONA exposes three operational primitives: • FIND -- discovers an entity or a data item by its name • Long-lived FINDs (ttl > 0) are subscriptions, which set up forwarding state • REGISTER -- advertises ownership of a data item • PUSH -- disseminates new content to all current subscribers • DHes propagate REGISTERrequests and route FINDs to nearby copies of data Data REGs FINDs
Load Shedding and Congestion Control Snapshot of Routing Table at DH D Snapshot of Congestion Table at DH D
Optimality of Load Distribution OPT_LS(Sk, Cn):“Given a set of clients and servers with specific query rates and capacities respectively, find an optimal mapping between clients and servers to minimize the total latency incurred by queries in the system”. c1 c2 c3.. cn-1 cn We can show that OPT_LS is NP-complete (reduction from the k-partition problem) S1 S2 S3 Sample construction for an instance of a 3-partition problem C={ a1,a2,a3….aN}; • Cap(S1)=Cap(S2)=Cap(S3)=D • L(S1)=L(S2)=L(S3)=1
Simulation Studies • Event-driven simulator written in Java • Observes average latency and CDFs of latency • Simulation parameters: • A 10 node topology using GT-ITM and manually generated topologies • Latency between links: Pre-Assigned (GT-ITM/manual)
OR OR appHeader != HTTP type == REGISTER ttl != 0 DONA Administrative Policies • Goal: ability to selectively deny incoming FIND and REGISTER requests based on fields in request header. • Example: DENY (type == REGISTER)OR (ttl != 0)OR (appHeader != HTTP) policy action Parse the expression, convert it to a tree • DH sends a DENY message to inform the client of its decision
Client Access Control • DONA provides a framework for implementing user access control mechanisms (authentication and authorization). • AUTHENTICATE policy action instructs a DH to request and verify client’s credentials prior to granting its request. AUTHENTICATE group1 Certif (type == FIND) authentication domain credentials format policy action First-hop DH Client node FIND Packet contain authToken Policy Module Passwd Certif Verifier 1 Verifier 2 Application Authentication Agent 8 DENY AUTH 5 4 1 6 7 10 credentials request data request FIND AUTHENTICATE ACCEPT 9 3 2 authToken DONA DONA
RSS Client 1 DH 4 Client Proxy 1 DH 1 DH 3 Client Proxy 2 DH 2 Server Proxy DONA RSS Client 2 RSS Source Services and Applications • RSS content dissemination on top of DONA: • Server-side proxy registers with a DH on behalf of content source. • Client-side proxies convert RSS refresh requests into long-lived FINDs (subscriptions), which set up forwarding state in the topology. • PUSH operation propagates content updates through the dissemination tree REGISTER PUSH FIND (ttl > 0) Reverse-path forwarding entry
DONA Services and Applications (cont.) • DONA rendezvous capabilities simplify design of Overlay Multicast: • Global multicast routing state = shared undirected tree • New group members use DONA to discover a nearby member and attach themselves to the tree • To transmit a packet to the group: • Sender locates a nearby member (M), forwards the packet to M • Each member forwards the packet to its neighbors on the tree M4 ATTACH SEND • DONA routing ensures that resulting trees are geographically efficient M2 M1 FIND FIND S1 sender REGISTER M5 M6