Networking 2012 On inter-domain name resolution for information-centric networks K.V. Katsaros, N. Fotiou, X. Vasilakos, C.N. Ververidis, C. Tsilopoulos,

Networking 2012

On inter-domain name resolution for information-centric networks

K.V. Katsaros, N. Fotiou, X. Vasilakos,

C.N. Ververidis, C. Tsilopoulos,

G. Xylomenos, and G. C. Polyzos

AUEB

Outline

• Name resolution in ICN• Related work• An enhanced DHT-based NRS• Performance evaluation & results• Conclusions

Name resolution in ICN

• ICN has many requirements from name resolution• Operation over a flat identifier namespace• Scalability

– Ability to efficiently handle 1015 names• Fault tolerance and fault isolation

– No centralized architecture• Low signaling overhead• Low latency (low response time)

– Efficient routing and load distribution• Routing policy compliance

Related work

• DHTs (Chord, Pastry)– Logarithmic number of hops required– Not compliant with routing policies

• MDHT and (old) PSIRP approach– Multilevel DHT with aggregation at higher levels– Questionable scalability and routing compliance

• DONA– Hierarchical aggregation of information– Requests are propagated upwards– Strictly follows customer-provider relationships

An enhanced DHT-based NRS

• H-Pastry: hierarchical DHT taking into account– Administrative domain boundaries – Inter-domain routing policies

• H-Pastry results in– Reduced path stretch (similar to regular Pastry)

• By 55% (Chord) and 47% (Crescendo)

– Confined traffic within administrative boundaries • 27% less inter-domain hops (Pastry)• 55% shorter intra-domain paths (Pastry)

– Reduced valley-free policy violations per path• By 56% (Chord), 31% (Pastry) and 36% (Crescendo)

An enhanced DHT-based NRS

• Outline of the H-Pastry based NRS

Performance evaluation & results

• Evaluation dimensions– Load: memory, signaling, processing overhead – Routing performance

• Particular attention paid to the effects of– Underlying network structure on performance– Popularity characteristics of content on caching

• Evaluation Setup: Topology– Scaled-down but realistic topology– 400 domains, 6-levels, peering and multihoming– Number of RV points: 4400

Performance evaluation & results

• Evaluation Setup: Workload– Original workload generated by GlobeTraff

• See our paper in IFIP NTMS 2012

– Focus on signaling to locate & start transmission– Data plane traffic mix translated to control plane

• Νumber of items for each traffic type • Data volume / median item size per traffic type

– Modeled popularity & temporal evolution• 25GB of total traffic

– ~2.5M subscriptions for ~1M objects

Performance evaluation & results

• Routing stretch in inter-domain hops– Ratio of DHT-NRS / DONA hops– Infinite Cache Size (ICS), m is the DONA median

• Caching performance– Works for popular items– 34% worse than DONA

Performance evaluation & results

• State: item entries per node• DHT-NRS considerably better than DONA

– Note the log scale for the x-axis!– DHT-NRS achieves better state distribution

Performance evaluation & results

• State distribution across hierarchy levels– Roughly 50% of access networks at level 2– DHT-NRS achieves a better state distribution– DONA is penalized by topology structure

Performance evaluation & results

• Lookup signaling overhead– Includes terminated + forwarded messages– DONA works better for most of the nodes– In DHT-NRS messages cross more nodes

Performance evaluation & results

• Lookup overhead distribution across hierarchy levels– DONA is problematic at level 1 (as expected)– DHT-NRS is also hit at the topmost level

• Subscribe/Notify messages often go through level 1

Performance evaluation & results

• Advertisement/registration signaling overhead– Inter-domain hop transmissions per registration– DHT-NRS requires 6.34 transmissions (at 0%m)– DONA requires 35.56 transmissions

• Excessive inter-domain traffic load for DONA – (Limited) flooding method for registrations– Multihoming plays a critical role

• Registrations sent to multiple higher levels• 56.75% of all domains are multi-homed• 2.4 providers on average for them

Conclusions

• Routing efficiency– Caching in DHT-NRS cannot compete with DONA– Stretch values range from 1.95 to 2.84

• Memory and lookup processing overhead– DHT-NRS considerably better– DONA has a highly skewed distribution

• Registration processing overhead– DONA is almost 6 times worse than DHT-NRS

• The problem with DONA is mainly localized– Large-scale centralized solutions (e.g. cloud)?