Movie Delivery in the Internet
- in 5 minutes Jorg Liebeherr
Department of Computer Science
University of Virginia
Interesting Issues
• Streaming vs. Download
• Storage
• VBR vs. CBR
• class-level vs. flow-level QoS
Media Types and their Demand
low
Movies
Live TV
Games
Bandwidth Demand
high
• have hardly real-time needs
• have hardly bandwidth needs
• High-bandwidth media types with stringent QoS demand:
• don’t exist in large numbers
• don’t exist
Interactive VR
Web-style
content
• Most media types:
Telephony
Control
online
(stringent)
QoS Demand
offline
(less stringent)
Unsubstantiated claim
• Claim:
For media delivery, most of
the bandwidth will be
consumed by “offline highbandwidth” media types
Other
Traditionally, are
handled via overprovisioning
or simple (TOS-like)
differentiation
Offline
highbandwidth
• Two ways to transmit these
media:
• Download
• Streaming
Download Time of an MPEG-2 Movie
• 90 min @ 6 Mbps  approx. 4 GB
100000
Probably not of interest
10000
Probably Streaming
1000
100
Download or Streaming
10
1
0.1
Available Bandwidth (Mbps)
10
00
00
0
00
10
00
00
10
00
10
0
10
10
1
0
Probably Download
0.01
0.
1
Download Time (seconds)
1000000
Duration of Movie
Download vs. Streaming
• QoS issues:
• probably none for downloads
in core network (due to high data
rates)
for resources)
Key Issue:
Placement of content at the edges of
the core network determines userlevel latencies.
cache
slow
(<stream
10 Mbps)
• probably none for downstream streaming in access
network (due to lack of contention
Access
Service
Provider
very fast
Core Network
download
(> 100 Gbps)
• Operating point will be
determined by available
technology
Cable Tree
Cable Modems
Stations
Summary: Movie delivery in the Internet within 5 years
• Classical QoS concepts may not be useful
• Storage issues seem more important  Good target for
QoS research
• Disclaimer: The above comments do not apply to media types with stringent
QoS demands:
• telephony
• games
• control
• interactive VR
Streaming Media: How to deal with VBR traffic?
• Add delay: Smoothing turns VBR into
CBR traffic
16000
14000
• Traffic theory: Statistical multiplexing
theory tells us that aggregated VBR
sources appear smooth
Traffic
12000
10000
8000
6000
4000
2000
0
0
•Economy of scale: If number of
receivers is large, peak rate allocation is
justifiable
100
200
300
400
500
Frame number
600
700
800
900
1000
Class-level vs. flow-level QoS
• This is a matter of desired effort of deployment and desired QoS:
• Theoretical understanding of flow-level QoS is very good, and
catching up for class-level QoS
• Class-level QoS is useful to protect legacy traffic (best effort, TCP)
• Problems (of both types of QoS):
• QoS notions are revamped before they mature
• The quality of commercial QoS implementations is insufficient