コンテンツ中心ネットワークに最適な協調ルーティングプロトコル
Improve the In-Network Caching Performance of
Content-Centric Networking Using a Cooperative Routing
Problems
Summary
o Shortest path routing (SP) is the default routing scheme of CCN
o By using the shortest paths, CCN routers may use disjoint paths
to forward their interest messages, even though they are
asking for the same content
o Each CCN router benefits from the cached content only by its
previously generated content requests
o Content Store (CS) of a CCN router must handle diverse
content objects when irrelevant data messages arrive at the
router
o Content objects may be so many that the CS cannot manage
them efficiently, which may result in a poor cache hit rate
We propose an optimal cooperative routing protocol (OCRP) for
the Content-Centric Networking (CCN) architecture, which
focuses on selecting the optimal routing path based on the content
retrieval statistics at a flow-level, to improve the in-network caching
utilization. The objective of OCRP is to selectively aggregate the
multiple flows of content requests onto the same path to improve
the cache utilization while mitigating the cache contention in the
Content Store (CS) of CCN routers. The optimal routing path is
obtained by binary linear optimization under flow conservation
constraint, cache contention mitigating constraint, and path length
constraint.
Our Proposed Method
Concept of Optimal Cooperative Routing
Overview of OCRP
o In CCN, content is defined by hierarchically constructed name
o Content objects originating from the same content source
share the same prefix
o We classify content flows based on the similarity of content
names by tracking popularly cited prefixes in Forwarding
Information Base (FIB)
o Similar content flows should travel on the same path
o Irrelevant content flows should travel on disjoint paths
Performance Evaluation
Simulation Setup
Network topology
The number of routers
Random mesh network
60 routers, degree=6
Content Store (CS) size
8, 16, 32
Content population
Popularity distribution
100 objects/prefix
Zipf distribution,
𝛂=0.2 to 2.0
The number of prefix groups
The number of requester routers in prefix group
Cache management scheme
Caching policy
Cache replacement policy
3
10
Probabilistic caching, Prob(p),
p ∈ {1.0, 0.5, 0.1, 0.01}
LFU, LRU, and RR
Performance versus caching decision and cache replacement policies
o Simulator: ndnSIM (an NS-3 based network simulator for NDN research)
o Evaluation metrics:
1) Server load
2) Round-trip hop distance of content retrieval
o Benchmark schemes:
1) Optimal Cooperative Routing Protocol (OCRP with k = 0, 10)
2) Shortest path routing (SP)
3) Cooperative routing protocol for CCN (CP with k = 0, 10):
• No limit of the prefix groups sharing the same link
• Irrelevant content flows may enter the same router
Performance versus content popularity distribution and CS size
Saran Tarnoi (サラン タノイ)
Yusheng Ji (計 宇生)
TEL : 03-4212-2661
TEL : 03-4212-2525
Email : [email protected]
Email : [email protected]