Computer Networks Lecture #8 PDF

Computer Networks Lecture #8 In the last lecture Control plane in SDN Internet Control Message Protocol (ICMP) Network management & con guration fi Today Information-centric networking Motivations Basic architecture Routing / Forwarding Motivations O...

Computer Networks Lecture #8 In the last lecture Control plane in SDN Internet Control Message Protocol (ICMP) Network management & con guration fi Today Information-centric networking Motivations Basic architecture Routing / Forwarding Motivations Observation #1 TCP/IP based Internet Designed at 1960s End-to-end communication model Location-based identi er : IP fi Observation #1 TCP/IP based Internet Designed at 1960s End-to-end communication model Location-based identi er : IP Internet tra c trend Massive amount of data Replicated data ffi fi Observation #2 Wired → Wireless Mobility becomes strong requirement The change of location (IP address) breaks existing connection Mobility management overhead is introduced Claim Let’s move on from (old, existing) IP to new network layer in which Data could be provided more e ciently Mobility could be more easily implemented Data security could be enhanced … ffi Information-Centric Networking (ICN) IP networking : request to where New networking : request for what Information-Centric Networking (ICN) From abstracted point of view, Internet is considered as a big Database Internet Information-Centric Networking (ICN) From abstracted point of view, Internet is considered as a big Database Internet ICN is an approach to access data distributed over the Internet Information-Centric Networking (ICN) Receiver-driven communication Requester initiates communication Queries (interests) / Data (responses) cf.) sender-driven communication in TCP/IP networking Content is addressed by its ID (e.g., name) rather than the location identi er (e.g. IP address) In-network caching can be performed at any network nodes Routers do more jobs cf.) routing / switching in TCP/IP networking fi ICN Architectures ICN proposals Centralized architecture Resolve Publish Local Local Pub/sub Internet routing paradigm Global NRS NRS NRS Locator Resolve Publish Flat name structure Locator GET Caches at the edge routers GET GET Source USER Data Object Data Object Name Resolution System Decentralized architecture Named data networking (NDN) Content centric networking (CCN) Named Data Networking (NDN) Content is addressed by its name Content name has a hierarchical structure. e.g., Youtube/movies/batman-III.mpg/I/… KNU/CSE/ComputerNetworks/Lecture1.avi/4/… Messages in NDN Query packets Called Interest Name: an identi er Nonce: guaranteeing uniqueness of the content combined with name Selector: matching & security relevant information Guider: relevant to routing / forwarding Data packets Called Response Name : an identi er MetaInfo Content: data Signature: security mechanism fi fi NDN Data Structures Forwarding Information Base (FIB) Similar to routing table Name pre xes are populated Used for routing interests Pending Interest Table (PIT) Store interests Mainly used for routing response Content Store (CS) Cache space for storing data fi Forwarding/Routing in NDN Forwarding/Routing in NDN Forwarding/Routing in NDN Forwarding/Routing in NDN Forwarding/Routing in NDN Forwarding/Routing in NDN NDN forwarding process summary Comparison: routing table Both IP and NDN have FIB IP - match destination IP with FIB entries to nd next-hop NDN - match interest name with FIB entries to nd next-hop The concept is the same IP Destination networks announce address pre xes Other routers compute best path NDN can do the same Content provider announce name pre xes But the protocol has to be di erent in NDN since routing messages need to Other routers compute paths be Interest/Data and multiple next-hops Can use the same routing algorithm, e.g., link state should be allowed fi ff fi fi fi Comparison: routing table NDN’s routing table size may be much bigger than IP’s IP routing table Cannot t all individual addresses in the table Aggregate individual addresses into pre xes Only allow pre xes in routing table In practice, only allow /24 or shorter pre xes globally Table size increases fast Multi-homing, mobility, tra c engineering, etc. are breaking aggregation fi fi ffi fi fi Comparison: routing table NDN routing table Cannot t all names in the table Only store name pre xes Operations may limit what pre xes can go into global FIB Use map-and-encap Map application name to routable name pre xes, which usually belong to ISPs ATT/nytimes/news/… fi fi fi fi Comparison: routing IP routing is fully responsible for the success of packet delivery Need to compute right paths and converge fast NDN routing is only a helper to forwarding Still need to populate FIB Doesn’t have to be perfect NDN allows new types of routing protocols Comparison: routing loop detection IP cannot detect looping packets IP routers do not keep tracks of each packet TTL is used to mitigate issues from routing loop NDN keep tracks of interests in PIT Each interest carries a random number : nonce Name + Nounce can di erentiate interests A looping interest would have a matching record in PIT, which allow looping packets to be successfully detected ff Comparison: failure detection IP cannot detect failure : one-way tra c & stateless NDN can observe data-plane performance Response is supposed to come back within reasonable time. If not? Failed If response returns, the next-hop works Can also record the time it takes to get the data If a PIT entry times out, something wrong might happen Drop due to congestion, link failure, loops, etc Interest NACK: explicitly returning interest to signal upstream problem ffi Comparison: multipath forwarding Multipath routing in IP is challenging Need to maintain routing consistency to make sure that there will be no routing loop It’s much easier in NDN Simply try a next-hop, and will know whether it works or not How to make use of the multiple forwarding options is the job of forwarding strategy, an important new feature enabled by NDN Comparison: adaptive forwarding Is enabled by named data, two-way tra c and datagram state Very e cient in handling link failure, congestion, and hijacks ffi ffi cf) Forwarding in NDN Request forwarding is performed hop-by-hop by every node Response is forwarded along the reverse path that Interest (request) has traversed How to exploit copies of item stored on o -path nodes? ff cf) How to use off-path cached content Control plain protocol To distribute information about mid-long term item copies Similar to the current routing protocol in IP (e.g., OSPF, …) Data plain approaches To exploit short-term content availability Exploitation Exploration Hybrid of the two : exploitation by default, optionally exploration for the popular content  Comparison: mobility IP has fundamental limitation Node address is bounded by location Delay, unreliable, costly to support mobility NDN inherently support customer’s mobility Majority of services are content retrieval from the xed node fi  Comparison: mobility IP has fundamental limitation Node address is bounded by location Delay, unreliable, costly to support mobility NDN inherently support customer’s mobility Majority of services are content retrieval from the xed node fi  Comparison: mobility IP has fundamental limitation Node address is bounded by location Delay, unreliable, costly to support mobility NDN inherently support customer’s mobility Majority of services are content retrieval from the xed node fi Comparison: mobility IP has fundamental limitation Node address is bounded by location Delay, unreliable, costly to support mobility Producer’s mobility Via rendezvous point (depot) Producer becomes another customer w.r.t rendezvous point Comparison: mobility Addressable entity Content in NDN vs. A location-speci c host in IP Underlying routing target Content in NDN vs a host in IP Interface for communication A socket bounded IP addresses is used vs. Content pub/sub interface is used Security A channel must be secured in IP vs. Content itself is secured independently fi Comparison: multihoming IP Host multi-homing in TCP/IP di cult (e.g., multipath TCP) Basically, TCP connection is created between two end interfaces NDN Does not depend on interface addresses Requests can be multiplexed over any interfaces Applications are hidden from this complexity → Never need to know interface addresses ffi Comparison: session / connection IP Majority of IP tra c is connection-oriented Congestion / ow control / reliability mechanism Mobility therefore requires TCP session maintenance NDN Congestion control / reliability can be achieved solely by consumer or in the hop-by-hop manner (not in the manner of end-to-end connection oriented) No need to exchange parameters fl ffi Comparison: resilience in dynamic topology IP TCP/IP is dependent on host availability (the existence of a designated host) Mobile networks are particularly vulnerable NDN Content could be anywhere in the network Content is not statically bound to locations Any sources can be used: ubiquitous caching & no single point of failure Comparison: abstraction of network addr. IP Some applications use network address Registering with BitTorrent tracker Necessitate a persistent address Or IP references can be stale NDN Detach application from network address Uses addresses that are already application-layer concept Comparison: inference scoping IP Information is often interpreted from host locations e.g., country, … NDN Makes an explicit split between content and location/user Not necessary to interpret information  Summary Simpli ed network management E cient usage of network resources Tra c reduction and localization Seamless connectivity Service exibility ffi ffi fi fl Questions?