A Na Native C Content Di Discover ery Mechan anism f for or NDN Onur Ascigil , Vasilis Sourlas, Ioannis Psaras, and George Pavlou Department of Electronic and Electrical Engineering, University College London, UK 1
Outlin line • Content discovery • Opportunistic Off-path Content Discovery • Forwarding Strategies • Results • Future Work &Conclusions 2
Cont ntent nt D Discovery 3
Content Discovery • Goal: Retrieve a nearby (ideally the nearest!) copy of the content • Difficult to achieve without significant ``overhead’’ in practice • Why? • Placement of Data into the Content Stores happens frequently • What does NDN/CCN do? • Route Interests to content origins • Search content opportunistically on-path (i.e., along the default path) • Existing Solutions for Content Discovery: • Opportunistic on-path • Coordinated off-path 4
Content Discovery • Opportunistic on-path: limited gain, without overhead • Coordinated off-path: coordination and communication overhead • Using control plane: Advertise content names • Using a function: A Hash function determines the placement and routing • … • What do we propose? • integrate an “opportunistic off-path content discovery mechanism” to the existing Interest/Data processing pipeline of NDN • With minimal changes to NDN packet processing • Without introducing excessive overhead 5
Opportunistic Of Off-path C Content Discovery 6
Opportunistic off-path Content Discovery • Satisfied Interest Table (SIT): Caches trails of Data packets FIB Prefix Next-hop /ndn T SIT SIT R Data Name Next-hop Name Next-hop /ndn/xyz H2 /ndn/xyz /ndn/xyz S T Interest … /ndn/xyz T U Interest FIB S /ndn/xyz Prefix Next-hop /ndn U FIB Prefix Next-hop 7 /ndn T H1
For orwarding St g Strategi egies es with F FIB & B & SI SIT 8
Forwarding Strategies with FIB & SIT • Breadcrumb SIT R Name Next-hop Name Next-hop H2 /ndn/xyz /ndn/xyz S, R S Wait in the PIT Interest Interest … CS /ndn/xyz /ndn/xyz U T FIB S /ndn/xyz Prefix Next-hop /ndn U Data /ndn/xyz 9 H1
Forwarding Strategies with FIB & SIT • Breadcrumb SIT R Name Next-hop Name Next-hop H2 /ndn/xyz S Wait in the PIT … Interest Interest CS /ndn/xyz /ndn/xyz T U FIB S Prefix Next-hop /ndn U NACK /ndn/xyz 10 H1
Forwarding Strategies with FIB & SIT • Multicast SIT R Data Name Next-hop Name Next-hop /ndn/xyz H2 /ndn/xyz /ndn/xyz S, R S Interest … Interest CS /ndn/xyz /ndn/xyz T U FIB S /ndn/xyz Prefix Next-hop /ndn U Data 11 /ndn/xyz H1
Forwarding Strategies with FIB & SIT • Multicast: Once forwarded downstream, an Interest follows a single SIT trail • Forwarding Strategy: Pick the freshest matching SIT entry SIT R Name Next-hop H2 /ndn/xyz S … Interest Interest CS /ndn/xyz /ndn/xyz T U FIB S Prefix Next-hop /ndn U 12 H1
Forwarding Strategies with FIB & SIT • Multicast: Once forwarded downstream, an Interest follows a single SIT trail • Forwarding Strategy: Pick the freshest matching SIT entry SIT R Name Next-hop Name Next-hop H2 /ndn/xyz S … Interest CS /ndn/xyz T U FIB S Prefix Next-hop /ndn U NACK /ndn/xyz 13 H1
Res esul ults - Multicast and Br Brea eadc dcrumb St Strategi egies es 14
Results- Settings • Topology : Rocketfuel ISP topology • Content Providers : Attached to 16 egress nodes are randomly chosen • Link Latencies: Penalty for leaving ISP network: extra 50-100 msec • Network Cache Capacity : %80 of content can be cached in the network • Request Rate : 100 Requests/sec (origins selected randomly) • Popularity Distribution : Zipf Parameter 0.7 • Experiment Duration: 1 Hour warm-up and 3 hours of experiment AS 3257 15
Results with Breadcrumb & Multicast Overlap 16
Results • Modified Multicast Strategy: • Each Interest is associated with a Forwarding Budget • spend the budget on: • sending a copy upstream (following FIB) • sending a copy downstream (following SIT) • spend it on both • Cost of sending Interest upstream/downstream • Static : Deduct one unit per each copy sent downstream. • Dynamic : Deduct variable amount per each copy sent downstream 17
Multicast with Forwarding Budget (Static) SIT Name Next- hops Q /x/y/z Y Interest: /x/y/z Interest: /x/y/z Interest: /x/y/z Budget = 5 Budget = 7 Budget = 6 R T S N Interest: /x/y/z Interest: /x/y/z OFF-PATH Y 18
Res esul ults - Multicas ast with F Forwar arding B g Budget 19
Res esul ults - Multicas ast with F Forwar arding B g Budget Satisfaction Overlap 20
Conclus usions ns • Opportunistic content discovery using SIT table • Significant increase in the percentage of requests satisfied from the caches • Forwarding strategies using SIT/FIB are introduced • Possible to limit the overhead with a Forwarding Budget and Dynamic Costs • Requires minimal changes to the packet processing of NDN • Security • End-user/host caches are not exploited by the mechanism • Exclude field can be used to ignore matching SIT entries • Our scheme inherits the existing cache poisoning problems with NDN 21
Thank you! Questions? 22
Recommend
More recommend
