Upload
lippo-group-digital
View
203
Download
4
Embed Size (px)
Citation preview
Caching in Information Centric Network (ICN)
Presented by: Priagung Khusumanegara
Outline
Introduction Characteristics of caching in ICN
Transparent Ubiquitous Fine Granularity
Caching optimization in ICN Dimensioning Decision policy
Challenges and future research2
Introduction
One of the important features of ICN is caching Speed up content distribution Improve network resource utilization
Characteristics of caching in ICN Transparent Ubiquitous Fine granularity3
Transparent
Making its routing and caching decisions on unified content names, essentially making these names network aware.
Several Challenges: Inconsistency between caching objective
ICN should make reasonable choice of its caching objective to balance between diverse traffic types.
Cross-application competitive sharing of cache space Different types of traffic differ significantly in their population
scale, object size and object popularity. ICN have to be able to efficiently share cache resources
between different traffic types. Line rate operation of caches
The cache management quite different from traditional disk-based management.4
Ubiquitous
Topology of the cache network evolves from hierarchical trees to arbitrary graphs.
ICN more dynamics because its general cache network topology, ubiquity of in network caches and volatility of cached content.
5
Traditional Caching System
Focus on Hierarchical Tree
Evolution
Information Centric Network
Focus on Arbitrary Graph
Fine Granularity
Different options for the granularity of caching: File-level
o Caching individual files as transmitted through the networko Typical size 1.5 Kb each – as proposed by CNN
Chunk-levelo Caching information chunkso An information object is split into a number of fixed-size
information chucks – as proposed by ICN
6
Fine Granularity (Cont’d)
Change of cache unit raises the following issues: Change of popularity
o Different chunks of a single file can have different access frequencies. Failure of independent reference assumption
o Traditional file-based caches are based on independent reference model
o Requests for different chunks of the same file are often correlated, e.g. in sequential order
Opportunity for more efficient use of the cache spaceo It possible to retrieve different parts of the same file from different
nodes, which speeds up the retrieval rate and improves the space utilization.
7
Techniques for ICN performance optimization
Focus on:– Cache dimensioning– Cache decision policy
8
Cache Dimensioning
- Since ICN cache should operate at line rate, the cache size that can be installed at each caching node is thus limited.
- There are two issues that remain to be addressed: How large the cache space should be to have noticeable
performance improvement?o Preferred to configure the cache size based on the router’s
performance disparity. How to allocate the storage resource across different cache nodes?
o Degree based allocation: the cache capacity allocated to a node is proportional to its node degree.9
Cache Decision Policy
Cache decision policy– It determines which objects are to be placed at
which cache nodes.– Two kinds of cache decision policy
Explicit cache coordination decision Implicit cache coordination
10
Explicit Cache Coordination
Object access pattern, cache network topology and each cache’s state as input for the calculation of the placement position of each object.
Common approaches can be classified into three categories:– Global – Path– Neighborhood
11
Explicit Cache Coordination (Cont’d)
Global coordination– Involves all cache nodes – Object placement based on network distance between
cache nodes and object access frequencies at each cache node
Path coordination– Only involves the cache nodes along the path from the
request hit place to the requesting client– e.g.: en-route web caching
12
Explicit Cache Coordination (Cont’d)
Neighborhood coordination– Coordination takes place among a node’s
neighborhood.– E.g.: Cooperative In-Network Caching (CINC)
13 Figure: The operation of coordination in network caching (CINC)
Implicit Cache Coordination
Each node does not need to know the state information of other cache nodes
LCE (leave copy everywhere) – Copy the object at each node along the
downloading path– Disadvantage: Degrade the performance of the
network and underutilize some of the network resources.
14
Implicit Cache Coordination (Cont’d)
Leave Copy Down (LCD):– When a cache hit occurs, this scheme only caches the
object at the direct downstream node
– Avoid a large number of copies of the same object
15
Figure: LCD (Leave Copy Down)
Implicit Cache Coordination (Cont’d)
Move Copy Down (MCD):– This scheme moves the object from the hit node
to its direct downstream node, and deletes the object from the hit node.
16 Figure: MCD (Move Copy Down)
Implicit Cache Coordination(Cont’d)
Copy with Probability- The requested object is copied with a given
probability p at each node along the returning path
17 Figure: Copy with Probability
Implicit Cache Coordination(Cont’d)
Random Copy One- The requested object is copied at one random
node along the returning path
18 Figure: Random Copy One
Implicit Cache Coordination(Cont’d)
Probability Cache- The requested object is copied at each node with
a probability. But, for each node, the probability varies.
19 Figure: Probability Cache
Correlation Between Cache Decisions
- WAVE adjusts the number of chunks cached at each node based on the file’s popularity
- When the number of requests for a file increases, WAVE reacts with exponential increase in the number of chucks cached for this file.
- A content router in WAVE explicitly sets the cache indication mark
- Once the chunk is cached, the cache indication mark is cleared.
20
Figure: Operation of WAVE
Challenges and Future Direction
Cache object popularity– Establish the chunk- level object popularity model from prior knowledge– Measure the chunk-level object popularity directly
Correlation between requests and correlation-based cache decision
– What is the inherent correlation between different requests, – How to model this correlation, and – How to optimize the cache decision policy based on the request
correlation,
ICN friendly network topology– What kind of network topology is suitable for ICN network.
21