《Hierarchical Caching Management for Software Defined
Content Network based on Node Value 》
Reporter：Jing Liu，China
Affiliation ：University of Science and Technology of China
Email：ljsmile@mail.ustc.edu.cn
Background
Users want the content instead of the IP address through the network
The original host-host model in IP is not suitable any more !
Background
Future Internet architecture:
CCN: Content-Centric Network
SDN :Software-Defined Network
Background
Why SDCN ?
The most existing CCN schemes are over the IP network
SDCN realizes CCN over SDN which does not rely on the IP
Protocol-Oblivious Forwarding (POF) is the extension of
OpenFlow proposed by HUA WEI.
SDCN over POF
Fig. SDCN Architecture
Node Value
How to describe the value of Content Node ?
node centricity metric
: tends to be the minimum path length
set a higher value for the center node which bears higher traffic
Node Value
Node centricity(Di ): the reciprocal of the sum of shortest path distance to other nodes
N is the number of nodes in network.
N
Di  1 /
d(i ,k )

k i
k 1, 
Normalization process of Di
N
D i （D i  N）/
Di

i
1
Node Value
Node connectivity ( Li ): the number of nodes connected to node i
Node Value
Node Value (Vi ) : Di    Li
Where  is the tradeoff between Node centricity and Node connectivity.
Hierarchical Cache Model
For S1: the sum of the shortest path :
2+1+3+2+2+3+4=17
Fig. Simple topology example
Hierarchical Cache Model
Table 1. Calculation of Node Value ( is 0.02)
Node
S1
Sum
17
Di
0.833422
Li
1
Li
0.02
Vi
0.853422
Rank
6
S2
17
0.833422
1
0.02
0.853422
7
S3
11
1.288015
4
0.08
1.368015
2
S4
16
0.885511
1
0.02
0.905511
5
S5
13
1.089859
2
0.04
1.129859
3
S6
10
1.416817
4
0.08
1.496817
1
S7
15
0.944545
2
0.04
0.984545
4
S8
20
0.708408
1
0.02
0.728408
8
Hierarchical Cache Model
The network with 8 nodes is divided into 3 layers.
The total cache C is divided equally into each layer and each layer further divides the C/3 equally
Table 2. The allocation of total cache space C
Cache size of Layer
nodes
Cache size of node
Layer1 (10%)
C/3
S6
C/3
Layer2 (30%)
C/3
S3, S5
C/6
Layer3 (60%)
C/3
S1,S2,S4,S7,S8
C/15
Hierarchical Cache Model
Fig The Hierarchical Model
Cache Decision Strategy : NVD
We assume the content popularity is known and set popularity threshold as T
Algorithm 1：Node Value based cache Decision Strategies NVD
1: Input
2:
Content: A data chunk D with popularity P
3: Begin
4: Check the value of each node along path
5:
if P > T
6:
Cache D on the node of V-max;
7:
else
8:
Cache D on the node of V-min;
9:
end if
10: End
Experiment and analysis
A topology with 50 nodes is built under Mininet.
The total storage budget is 600 objects, and 1000 files.
Content popularity is modelled with a Zipf distribution
Content requests are modelled as Poisson process.
The cache replacement policy is based on popularity
Experiment and analysis
P_hit = hit / request
Fig. The average P_hit of each layer in four scenarios
Experiment and analysis
Fig. P_hit of each node using LCE_Model vs NVD_Model
Conclusion
The proposed hierarchical cache model and cache policy can improve the
performance of network caching, including hit ratio and average hops of each request.
For further research, the study of node value considering more information
such as link bandwidth will be required.
The End