Loopback: Exploiting Collaborative Caches for Large-Scale Streaming

1
Loopback Exploiting Collaborative Cachesfor
Large-Scale Streaming

• Ewa Kusmierek, Yingfei Dong, Member, IEEE, and
  David H. C. Du, Fellow, IEEE

2
Outline

• Abstract
• Related work
• Client collaboration with loopback
• Loopback analytical model
• Local repair mechanism enhancing reliability
• Conclusion and future work

3
Abstract(1/2)

• Two-level streaming architecture
• content delivery network(CDN) to deliver video
  from central server to proxy servers.
• Proxy server delivers video with the help of
  client caches.
• Design Feature
• Loopback approach
• Local repair scheme

4
Abstract(2/2)

• Objective
• Reduce the required network bandwidth
• Reduce load of central server
• Reduce cache space of a proxy
• Address client failure problem

5
Outline

• Abstract
• Related work
• Client collaboration with loopback
• Loopback analytical model
• Local repair mechanism enhancing reliability
• Conclusion and future work

6
Related work

• P2Cast
• A session is formed by clients arriving close in
  time.
• Application-level forwarding tree.

Server
peer
peer
peer
peer
peer
7
Related work

• CDN-P2P hybrid architecture
• Divide data into fractions
• A Client may receive video stream from multiple
  peers, A client need to cache an entire video
• Client needs to caches an entire video

server
peer
peer
peer
8
Outline

• Abstract
• Related work
• Client collaboration with loopback
• Loopback analytical model
• Local repair mechanism enhancing reliability
• Conclusion and future work

9
Basic assumption for client

• Each client dynamically caches a portion of a
  video and storage space is limited
• Client delivers only one stream at a time only
  during its own video playback and for a short
  period of time after the playback ends
• Client may fail or choose to leave while
  delivering the video data to its peers.

10
Basic assumption for proxy

• Storage space is limited.
• Bandwidth is limited.
• The prefix of a video is cached by proxy server.

11
Forwarding Ring(1/3)

• Clients arriving close to each other in time form
  a forwarding ring
• First client receiving data from a proxy.
• Last client returning data to the proxy.
• First client receives the video prefix from the
  proxy and the remaining portion of a video from
  the central server

12
Forwarding Ring(2/3)

• Next client join on time
• Streamed to the newcomer.
• The frames that have been already
• transmitted are removed from the buffer.
• If next request arrive not in time
• Oldest frames are passed back to the proxy and
  evicted from the buffer.
• The late newcomer starts a new loop.

13
Forwarding Ring(3/3)

• Proxy does not maintain a copy of a frame after
  transmitting to a client.
• If the demand is high
• There are few long loops containing many clients.
• The entire video may be cached by the clients.
• Proxy only need to forward one stream to each
  loop and receive one stream from each loop

14
(No Transcript)
15
Outline

• Abstract
• Related work
• Client collaboration with loopback
• Loopback analytical model
• Local repair mechanism enhancing reliability
• Conclusion and future work

16
Loopback analytical model

• Analyze the resource usage at the proxy and the
  central server load due to a single video under a
  given client arrival process.
• Notation definition
• buffer size at each client
• arrival time of the i th client.
• storage space of the proxy( 0lt lt1)

17
Aggregate Loop Buffer Space
18
Data available locally
19
Proxy Buffer Space Utilization
20
Proxy I/O bandwidth usage
21
Central server load
22
Outline

• Abstract
• Related work
• Client collaboration with loopback
• Loopback analytical model
• Local repair mechanism enhancing reliability
• Conclusion and future work

23

• Client failure effect
• Loss data has to be obtained from central server,
  incurring delays.
• May affect succeeding clients in a loop.
• The higher the demand, the larger the influence
  of a failure on the performance
• Address this issue with redundant caching
  schemes.
• ?significantly reduces server load
• ? shortens the repairing delay caused by
  transmitting missing data

24
Complete-local and partial-local repair
25
Additional loads saved by local repairs
26
(No Transcript)
27
Outline

• Abstract
• Related work
• Client collaboration with loopback
• Loopback analytical model
• Loopback performance for multiple videos
• Local repair mechanism enhancing reliability
• Conclusion and future work

28
Conclusion

• Loopback mechanism for exploiting
• client collaboration in a two-level
• video streaming architecture.
• Improve resource usage
• Server Network bandwidth and I/O bandwidth
• Proxy Network bandwidth and I/O bandwidth
• Proxy storage space
• Analyze the effect of client failures and
  developed local repair approaches

29
Future work

• Allow varying amount of resources committed by
  each client
• Each client can specify how much disk space can
  be utilized
• According to network bandwidth , each client can
  decide how many clients he want to serve , and
  for what period of time

30
Central Server
Proxy
Proxy
Proxy
Peer
Peer
Peer
Peer
Peer
Peer