Popularity-based Caching in Content Centric Networks Thomas SILVERSTON Université de Lorraine Loria, CNRS UMR 7503 Inria Nancy – Grand Est 20 Mars 2014
Popularity-based Caching in Content Centric Networks
Thomas SILVERSTON
Université de Lorraine Loria, CNRS UMR 7503 Inria Nancy – Grand Est
20 Mars 2014
Mo#va#on § Internet is mostly used to access content
– Video: 86% of global consumer traffic by 2016 • Cisco VNI 2012
§ Internet: host-‐to-‐host communica#on – Users are interested with content, not loca#on
§ Communica#on Models – Client/Server – Content Delivery Networks (CDN) – Peer-‐to-‐Peer (P2P) 1
Communica#on Models: Client/Server
Server Client
ISP 4
DT
ISP 1
ISP 3
ISP 2
• Limited Resources of Server
• Not scalable to a large audience
2
ISP 4
DT
ISP 1
ISP 3
ISP 2
Server Client
Communica#on Models: Content Distribu#on Networks
CDN
CDN
CDN
• Replicated Server
• Scalable
• Costly
3
Communica#on Models: P2P
Server Client
ISP 4
DT
ISP 1
ISP 3
ISP 2
• Client and/or server
• Scalable
• Traffic
4
Mo#va#on
§ Internet is mostly used to access content – Video: 86% of global consumer traffic by 2016
• Cisco VNI 2012
§ Internet: host-‐to-‐host communica#on
§ Users are interested with content, not loca#on
§ Future Internet: Informa#on Centric Networks – Content Centric Networks (CCN)
• Van Jacobson [ACM CoNext 2009] 5
CCN Overview § In-‐Network Caching
§ Packet address refers to content and not loca#on
§ Two primi#ves – Interest, user requests content by issuing an Interest message
– Data, a node having the content answer with a Data message
§ Data at the core of the communica#on
§ ‘Network Layer’ for Content Delivery 6
CCN Overview
§ Host 1 wants content /content/abc.flv
§ Host 1 issues an Interest message
7
CCN Overview
§ CCN Node has /content/abc.flv in its Cache
§ CCN Node issues a Data message
8
CCN Overview
§ CCN Node has NOT /content/abc.flv in its cache
§ CCN Node transfers the Interest to the rest of the Network
9
CCN Overview
§ /content/abc.flv is found
§ A Data message is transmi\ed along the delivery Path
§ CCN makes room by evic#ng some content
§ CCN Node stores /content/abc.flv and forward it to Host 1 10
CCN Cache Management § Caching along the delivery Path
– In-‐Network caching
§ Replacement Policies – Decide the element to be replaced
• LRU, FIFO, MRU MFU etc. • Well-‐studied for systems, memory etc.
§ Caching Strategy – Decide whether to cache content
• Huge cache, Fricker et. al [IEEE NOMEN 2012] • Cache less, Chai et. al, [IFIP Networking 2012] 11
Essential to design caching management for CCN
Popularity-‐based Caching Strategy § CCN in-‐network caching
– Storing content at every nodes on the path – Overloading nodes and network Resources
§ MPC Caching Strategy [IEEE ICC 2013] – Decide to cache the content only if the content is popular • Counts locally #Interests (Requests) for a Content • #Interests > Popularity Threshold➡Popular
12
Case Study: CCN
13
A B
C
E e
Scenario
A requests e
B requests d A requests d
C requests d d
D
Case Study: CCN
14
B
C
E d e
e
e
e
d
d d
e
d
X 3
X 3
#Replications
A
D
Case Study: MPC
15
B
C
E d e
Scenario
A requests e
B requests d A requests d
C requests d
A
D
Case Study: MPC
16
B
C
E d e
d
d
d
X 0
X 2
e #Replications
A
D
Content Popularity-‐based Caching
§ CCN in path caching – 3 replicas of each content d and e
§ MPC caches only popular content – 2 replicas of content d (popular) – No replica of content e – Pro-‐ac#ve caching of popular content
17
Evalua#on
§ Does Caching only popular content improve CCN performances?
§ ccnSim simulator – Standard parameters
§ Comparison – MPC vs. CCN
18
MPC Evalua#on
19
0.6
0.8
1
Tree Abilene Tiger2 GeantDTelekomLevel3
Cach
e H
it R
atio
Topologies
CCN (LRU+Always)MPC
MPC Evalua#on
20
0
0.5
1
Tree Abilene Tiger2 Geant DTelekom Level3
Ratio
of C
ache
d El
emen
ts
Topologies
CCN (LRU+Always)MPC
Popularity
1. Content
2. Users
21
Future Internet
§ Internet has become a « content network » – Video counts for 86% of traffic up to 2016
§ Users 'ac#vity: Online Social Network – Internet has become a « social network »
§ Future Internet – OSN over CCN
22
Future Internet OSN over CCN
23
Future Internet
§ Internet has become a « content network »
§ Users 'ac#vity: Online Social Network – Internet has become a « social network »
§ Future Internet – OSN over CCN
§ Does OSN may improve CCN ? – Privilege content published by popular users
§ Does popularity of users help In CCN?
24
Popularity of Users
§ Score for users regarding their importance
§ Eigenvector centrality measures
§ PageRank centrality measures – Google
§ If score(user) > Avg_score – Influen#als <-‐ user
25
Simula#on Environment
26
Simula#on Environment § Social Networks
– LastFM and Facebook data set
§ Network Topology – iNet (default parameters) : 3,037 CCN nodes
§ Ac#vity Traces – Sonetor: SOcial NEtwork Traffic generatOR – ~50,000 ac#vi#es by synthe#c traces [IEEE ICC 2014]
27
Users’ Interac#on Model
28
Users Popularity-‐based Caching
29
Sacs/PageRank Sacs/Eigenvector CCN
Users Popularity-‐based Caching
30
Sacs/PageRank Sacs/Eigenvector CCN
PlanetLab Experiments
31
PlanetLab Experiments
32
Summary § Internet: a “Content” and a “Social” network
§ CCN as a novel “Network Layer” for Content Delivery in the Future Internet
– In-‐network Caching
§ Popularity-‐based Caching Strategies for CCN – Content Popularity – Users Popularity [IFIP Networking 2014]
§ Improve Cache Performances
§ Experiments on CCNx and PlanetLab 33
Sorware-‐Defined Networking § Decoupling
– Data Plane (traffic forwarding) from – Control Plane (Rou#ng decision)
§ Controller – Openflow
§ ICN/CCN: alterna#ves to TCP/IP architecture – Focus on content and not hosts – Lack of deployments
34
SDN integra#on with ICN
§ Openflow-‐compliant ICN – CONET architecture – IP-‐compa#ble layer to enable ICN nodes
§ Abstrac#on layer between CCN nodes and Controllers based on IP
– Not full ICN
§ Goal: SDN with ICN without IP
35
36
SDN integra#on with CCN
§ Model this SDN/CCN architecture
§ Simula#ons with NS-‐3
§ Also with Controllers (SDN): – Network security (Firewall, Access Control), etc.
37
Ongoing Work
Thank You!
Ques#ons?
38
39