Proposed Caching Management Scheme

Report
Authors: Jason Min Wang, Brahim Bensaou
Publisher: GLOBECOM 2012
Presenter: Chai-Yi Chu
Date: 2013/05/08
1


Introduction
Proposed Caching Management Scheme
◦ Caching Decision Policy
◦ Replacement Strategies

Simulation
◦ Experimental Methodology
◦ Experiment Results
2

propose a new caching scheme for such CCN networks
and evaluate the in-network caching performance of
this policy by comparing it with that of the default
proposed policy via simulation.
3
Characteristics that have crucial influence on the
caching performance
1. Locality of references
2. Content popularity distribution
3. One-time referencing
4. Heavily-tailed object size distribution

4

Caching Decision Policy
◦ Resemblance to the LCD algorithm (Leave Copy Down)
◦ Choosing the immediate downstream node of the cache hit
point as the primary candidate place to replicate the data
packet.
5
6
◦  : the number of interfaces saved in the PIT entry, that is,
from how many distinct interfaces requests for the same
namedchunk  are aggregated.
◦  : the actual number of individual requests for p at an edge
node.
7
8

Replacement Strategies
◦ Edge nodes
 A modification of the Greedy Dual-size algorithm.
 Each cached chunk of data  is associated with a value  .
  : the hop count needed to fetch the packet.
 An “inflation” value  = min  .

9
10
◦ Intermediate nodes
 Each cached chunk of data  is associated with a value  .
 Interface .
1
2
 Diversity information will be recorded in  and  is used
to leave breadcrumbs on the access statistics of  after it has
been cached.
11
12

Implemented a simplified CCN model on top of
Omnet++
◦ simulation model includes three basic components of CCN
i.e., CS, PIT and FIB
◦ other features of CCN (e.g., hierarchical naming, routing,
security issues and so on) are not taken into account.
13

Experimental Methodology
◦ Network topology
14
◦ Workloads
 The synthetic Web workload generator ProWGen is used to
generate workloads for the two content servers.
15
◦ Performance metric
 systematic hit gain  =
ℎ∈ ℎ
∗ ℎ /
∈,∈ 
∗ 
 ℎ : the distance between node  and the original content server.
 ℎ : the amount of pending requests at edge nodes for the hitting
data.
  : the size of object  (chunks).
  : the hop distance between node  and the original content
server  of object .
 The closer the value of G is to 1, the better the in-network
caching system performs.
16
◦ Methodology
 cache size
 varied uniformly from 100 to 8,000 chunks for all nodes.
 The chunk size is set 10KB
 request aggregation
 request aggregation time can change the observed access pattern
and thus impact the hit rates of the nodes.
 cache management scheme
1.
2.
alwayscache+LRU (the initial proposal of CCN
proposed PCP+heterogeneous replacement algorithms
17

Impacts of cache size and content popularity
18
19

Impact of request aggregation
20

similar documents