zhen feng, mingwei xu, yu wang and qing li tsinghua university, beijing, china, globalcom2013 –...

17
Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR CACHE CONSISTENCY SUPPORT IN INFORMATION CENTRIC NETWORKING

Upload: alvin-todd

Post on 11-Jan-2016

220 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

Zhen Feng, Mingwei Xu, Yu Wang and Qing LiTsinghua University, Beijing, China,Globalcom2013 – NGN Symposium

Katto LabHiroto Kisara

AN ARCHITECTURE FOR CACHE CONSISTENCY SUPPORT IN INFORMATION CENTRIC NETWORKING

Page 2: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

2

Outline• Introduction• Related Work• Design of CUVS• Why traditional approach in ICN

• The hash tree structure

• The strong consistency of metadata

• The process of CUVS

• Evaluation and Result• Evaluation Setting

• Results• Infinite Cache Size Scenario

• Finite Cache Size Scenario

• Summary

• Conclusion

Page 3: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

3

Introduction• ICN is an architectural redesign proposed to deal

with the most prominent requirements.• Many schemes have been proposed. =>

caching

• when the origin copy of content changes sometime, the cache consistency among all the copies needs to be taken into consideration.

Contents

Contentswill be never

modified.

Contentswill be updated

in real time.

ContentsMay change occasionally

Page 4: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

4

Introduction (cont.)• no academic efforts have been put on the cache

consistency issue in the ICN field.• In author’s opinion, the wide spread caching in

ICN makes the issue more intractable.

And the traditional approaches which mainly aim to operate in a small size or relatively self-managed environment cannot be efficiently applied in ICN.

Content Update Validation System(CUVS)

Page 5: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

5

Introduction (cont.)Content Update Validation System(CUVS)• is an overlay system build upon the ICN network

layer.• provides the service of validating whether the

version number in a content name is the latest.• employs a hierarchical structure similar to DNS.

Evaluation• A testbed to evaluate their scheme is built and

implemented the NDN design.• The performance and overhead of four scheme:

1) Fixed TTL 2) Polling-Every-Time 3) Lease 4) CUVS

Page 6: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

6

Related WorkTraditional cache consistency schemes can be categorized into cache validation (Polling-Every-Time, TTL) and server invalidation (Lease).• TTL approach

When the TTL of a cached content is expired, the cache will send validation to the origin server.

• Polling-Every-TimeEach time a request is received, the cache will send validation to the origin server.

• Lease approachThe origin server will keep track of each cache which is called lease for a period of time. It will notify the caches whose lease is still valid. The caches have to update their lease to the server periodically.

The authors optimize this approach to provide a strong consistency

mechanism.

Page 7: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

7

Design of CUVS- Why traditional approaches fail in ICN

In the cache validation approach and the server invalidation approach, the origin server has two same roles:

1) serving contents 2) validating or invalidating caches

All the network devices may cache contents, so the validation requests from widespread caches may impact the performance of the origin server or even overwhelm it.

It is feasible in traditional network since the number of contents is limited and most importantly they behave in a

controlled manner.

In ICN, the situation has dramatically changed!!

Page 8: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

8

Design of CUVS- Why traditional approaches fail in ICN

Proposed idea originates from the separation of the two roles of the origin server.

1) The cache validation or invalidation functionThis function can be done by CUVS system which employs the wide spread servers located in every domain.

2) function of serving contentsThe origin server has only this function.

This improves the scalability and robustness of the ICN network.

Page 9: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

9

Design of CUVS- The hash tree structure

The advantage of hash tree structure is that you do not need to validate every piece of content in a name space.

If the root hash key of the name space is not changed, all of its contents remain unchanged.

The hash tree is a metadata of a name space, and it is generated by the origin server.

Whenever the content is updated, its hash tree should be recalculated.

Fig.1. Hash tree structure

Page 10: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

10

Design of CUVS- The strong consistency of metadata

Whether the content is stale or not can be decided by the CUVS servers according to the metadata.

Xin Chen proposed a strong consistency mechanism in DNS.

In their design, the local DNS server will use dynamic Lease approach to keep its record the same as targeted DNS server.

If there is an update, the targeted DNS server will notify every local DNS server to renew its cache

The authors apply this design to their proposed

scheme.

Fig.2. Website hash tree update multicast

Page 11: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

11

Design of CUVS- The process of CUVS

Fig.3. The working process of CUVS

Page 12: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

12

Evaluation and Results

Fig.4. Experimental topology

Performance

The average response time

OverheadThe origin server load

Bandwidth consumptionStaleness ratio

Page 13: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

13

Evaluation SettingTwo Scenarios:

・ Infinite cache size

・ Finite cache size

Size:

・ Interest and Data (with no content) packet: 20B

・ Data(with content) packet: 1KB

The request Interval:

・ Based on Poisson distribution

・ the mean value is 2s

The popularity:

Based on Zipf distribution

Duration of experiment:

10minutes

Content update interval:

・ infinite scenario: every 80s

・ finite scenario: divided into 3

grades

1) most popular 10 contents

never change

2) following 50 contents

every 40s

3) the other 40 contents

every 80s

Page 14: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

14

Results – Infinite Cache Size Scenario

Fig5(a) Staleness Ratio Fig5(b) Average response time

Fig5(c) Origin serever load Fig5(d) Bandwidth consumption

Page 15: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

15

Results – Finite Cache Size Scenario

Fig6(a) Average response time

Fig6(c) Bandwidth consumption

Fig6(b) Origin server load

Fig6(d) Staleness ratio

Page 16: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

16

Summary• Polling-Every-Time

• Lease

• Fixed TTL and CUVS

Advantage

Strong cache consistency

disadvantage

Long response timeHeavier load of origin

server

Advantage

Strong cache consistencyShortest response time

disadvantage

More bandwidth consumption

Heavier load of origin server

disadvantage

Weak cache consistency

Performance

TTL < CUVS

Advantage of CUVS

The load of the origin server is very low.

Page 17: Zhen Feng, Mingwei Xu, Yu Wang and Qing Li Tsinghua University, Beijing, China, Globalcom2013 – NGN Symposium Katto Lab Hiroto Kisara AN ARCHITECTURE FOR

17

Conclusion• In this paper, a cache consistency architecture named

Content Update Validation System (CUVS) is proposed. It aims to resolve the ubiquitous caching validation requirements in the ICN network. It is designed as a hierarchical overlay in ICN and provides the service of validating cached contents.

• The architecture has the following advantages.1) it relieves the origin server from the work of cache

validation.2) its validation service can be invoked by any routers in

the network3) its overlay architecture makes its design independent

from the preliminary ICN network layer designs and can be easily deployed.

4) the hierarchical structure of the system ensures good scalability.