NDN Live Video Broadcasting over Wireless LAN Menghan Li, Dan Pei, - - PowerPoint PPT Presentation

ndn live video broadcasting over wireless lan
SMART_READER_LITE
LIVE PREVIEW

NDN Live Video Broadcasting over Wireless LAN Menghan Li, Dan Pei, - - PowerPoint PPT Presentation

NDN Live Video Broadcasting over Wireless LAN Menghan Li, Dan Pei, Xiaoping Zhang, Ke Xu Tsinghua University Beichuan Zhang University of Arizona IP: Delivering packets to endpoints 0 4 8 16 19 31 TOS Length Version HLen Ident


slide-1
SLIDE 1

NDN Live Video Broadcasting

  • ver Wireless LAN

Menghan Li, Dan Pei, Xiaoping Zhang, Ke Xu Tsinghua University Beichuan Zhang University of Arizona

slide-2
SLIDE 2

IP: Delivering packets to endpoints

Version HLen TOS Length Ident Flags Offset TTL Protocol Checksum SourceAddr DestinationAddr Options (variable) Pad (variable) 4 8 16 19 31 Data

Source Address Destination Address

1

slide-3
SLIDE 3

Retrieving Named Data from the network

Named Data Networking (NDN)

Content Name Content Name

2

slide-4
SLIDE 4

IP’s Node Model

3

IP Packets

FIB Self?

Accept Forward

✗ lookup miss ✓ lookup hit ✗

Drop

✗ ✓ ✓

One-way traffic, stateless, no storage.

slide-5
SLIDE 5

NDN’s node model

Two-way traffic, stateful, explicit storage.

4 Content Store Pending Interest Table (PIT)

Data

forward Discard cache

✗ lookup miss ✓ lookup hit

Content Store Pending Interest Table (PIT) FIB

Interest

✗ ✓ ✗

forward

Data

Add Incoming Face

Drop or NACK Downstream Upstream

slide-6
SLIDE 6

Content Distribution Example

ISP ISP 36.7M Views

IP needs large infrastructure and complicated technical solutions.

5

slide-7
SLIDE 7

Content Distribution Example

ISP1 ISP3 ISP2 ISP4

36.7M Views

NDN’s multicast and caching are native and built-in.

6

slide-8
SLIDE 8

The problem at the last hop

Increasingly over wireless broadcast medium, ie, WiFi. But current NDN implementation treats it as multiple unicast tunnels between clients and access point (AP).

7 Video Server AP Internet AP AP AP

slide-9
SLIDE 9

Design Goals

Efficient and Scalable NDN-based live video broadcasting over WiFi

  • Support a good number of clients watching the same

live video streaming via the same WiFi AP at the same time. No modification to the MAC layer

  • WiFi doesn’t support broadcast or multicast very well.
  • Built our system based on NDN/UDP.

8

slide-10
SLIDE 10

Design Overview

Regulate Interests so that ideally only one Interest is sent to retrieve one video packet regardless of the number of clients.

  • One Interest unicast to the AP, which will forward it to

the video server.

  • One video data packet comes back to the AP, which

broadcasts the packet to all clients. How to regulate the generation of Interests? How to regulate the transmission of Interests?

9

slide-11
SLIDE 11

Regulating Interest Generation

Video Server AP Client NDN Testbed

Maintain an Interest sending window (W) at the client.

  • For pipelining multiple Interests to retrieve multiple data.
  • It needs to match data production at the server.

Client periodically asks for the latest data name to infer the rate of data production, and use that to set W.

W = RTT *(S2 − S1) / (T2 −T

1)

10

slide-12
SLIDE 12

Regulating Interest Transmission

Goal: suppress the transmission of duplicate Interests from different clients.

  • Ideally only one Interest is sent (from the Leader) to

retrieve one video packet. The AP chooses one client (e.g., the first one) to be the Leader, and others are Followers.

  • The Leader transmits Interests without delay.
  • The Followers delay the transmission of their Interests.
  • The AP periodically broadcasts a heartbeat packet to all

clients, stating who the Leader is, the current number of Followers, and the current progress of video data retrieval.

11

slide-13
SLIDE 13

Interest suppression at Followers

Outgoing Interests are delayed and stored in Delayed Interest Table (DIT).

  • Consumed when corresponding data is received.
  • Transmitted by certain probability when timeout or

packet loss is inferred. Inferring packet loss from data name or AP’s heartbeat message.

  • Significant out-of-order packets

The probability to transmit

  • N is the number of clients, and R is the number of times

that the application has been asking.

12

Pf =1/ (N − R)

slide-14
SLIDE 14

Keep track of clients

AP maintains who’s the Leader and the number of Followers. Every client periodically sends an Interest to learn the latest available data name.

  • This Interest is not delayed.
  • AP uses this Interest to keep track of the clients.

If a Follower leaves, AP just updates the count. If the Leader leaves, AP will assign the next client as Leader.

  • In absence of Leader sending Interests, clients will timeout

and send Interests on their own, based on a probability.

13

slide-15
SLIDE 15

Evaluation

Compare with two alternatives Unicast: The clients and AP transmit all NDN packets

  • ver UDP unicast tunnels.

Broadcast: All clients send all Interests over UDP unicast, and the AP broadcasts the returned Data to all the clients.

14

slide-16
SLIDE 16

Performance metrics

Buffering Rate: the number of buffering events per second. Buffering Ratio: the percentage of time spend in buffering. Interest Redundancy: the number of Interests sent vs. the number Data received. Data Redundancy: Data sent by AP vs. received by clients.

15

slide-17
SLIDE 17

Experiment setups

AP: commodity home router

  • 5GHz WiFi, 6Mbps broadcast bandwidth,

720MHz CPU 20 Clients

  • Ubuntu VM running a single client app with a

dedicated WiFi card. Software:

  • NDNVideo: 1Mbps H.264/AVC video

16

slide-18
SLIDE 18

Scalability to the number of clients

1 5 10 15 20 0.02 0.04 0.06 #client Buffer Rate(#/s)

NLB Ucast Bcast

0.026 0.01 0.041

1 5 10 15 20 10 20 30 40 50 #client Buffer Ratio(%)

NLB Ucast Bcast

14.03 4.02 1.99

NLB supported 20 clients at ease, while unicast and broadcast schemes showed clear limit.

17

slide-19
SLIDE 19

Packet redundancy

1 5 10 15 20 100 200 500 1000 #client

Interest Redundancy(%,log scale)

NLB Ucast Bcast Ideal Ucast

122.96

1 5 10 15 20 100 200 500 1000 #client

Data Redundancy(%,log scale)

NLB Ucast Bcast Ideal Ucast

118.75

NLB is able to suppress the majority of redundant packets, Interest and Data.

18

slide-20
SLIDE 20

AP’s CPU usage

1 5 10 15 20 20 40 60 80 100 #client CPU Usage(%)

NLB Ucast Bcast UCast: may be limited by CPU cycles Bcast: limited by 6Mbps broadcast bandwidth NLB: the total number of packets transmitted over WLAN grows very slowly 19

slide-21
SLIDE 21

Summary

NLB is a scalable and efficient receiver-driven broadcasting mechanism in wireless medium. It complements NDN’s multicast and caching capability in the wired Internet. It runs on commodity access points and mobile devices without any change to the MAC layer

20

slide-22
SLIDE 22

For More Information about NDN

http://www.named-data.net/

21