Better Than Best Effort

1
Better Than Best Effort?

• Bernd Girod
• Information Systems Laboratory
• Stanford University

2
Clean Slate, Inc.

• CleanSlate is a gang violence recovery, rage
  resolution and tattoo removal program that helps
  to erase the visible signs of gang involvement
  while working with the gang members to get back
  in touch with their emotions.  The gang culture
  does not allow for expression of perceived
  weakness such as sadness, vulnerability, fear and
  tears.  We give them permission to express these
  feelings through a Gang Recovery Group that meets
  Friday's on a weekly basis.  They are then
  connected with an individual counselor who gives
  them one-on-one support.  We work closely with
  the Southern California Counseling Center where
  the Gang Recovery Groups take place.  

3
Outline of the Talk

• QoS vs. best effort
• Rigid vs. elastic applications
• Link sharing
• Error control and delay
• Path diversity
• Suggestions for a New Internet

4
QoS vs. Best Effort

• Reservation-ism
• Real-time applications (voice and video) need
  bandwidth guarantees
• Real-time applications need packet loss
  guarantees
• Real-time applications need guaranteed delay
  bounds
• Best effort is fine for data applications
• Best Effort-ism
• Best Effort good enough for all applications
• Real-time applications can be made adaptive to
  cope with any level of service
• Overprovisioning always solves the problem, and
  its cheaper than QoS guarantees

5
Simple Model of A Shared Link

• Link of capacity C is shared among k flows
• Fair sharing each flow uses data rate C/k
• Homogeneous flows with same utility function
• Total utility

C
Breslau, Shenker, 1998
6
Rigid Applications
u

• Rigid applicationNo utility u below of minimum
  bit-rate B
• Maximum total utility Uk is achieved by
  admitting at most k flows

1
C/k
B
Breslau, Shenker, 1998
7
Rigid Applications (cont.)

• Expected loss in total utility w/o admission
  control
• Gap DU is substantial when number of admissable
  flows k is small
• Gap DU usually disappears with growing capacity
  C, but not for exceptionally heavy-tailed
  distributions Prk

Breslau, Shenker, 1998
8
Elastic Applications

• Elastic applications utility function u(k), such
  that total utility U(k)ku(C/k) increases with k
• Example u(C/k)1-aC/k
• All flows should be admitted best effort

u
C/k
9
Video Compression

• H.264 video coding for 2 different testsequences

• Video is elastic application
• Rate must be adapted to network throughput
• Rate control for live encoding
• Scalable representation or transcoding for stored
  content
• Utility function depends on content should use
  unequal rate allocation

Foreman Mobile
10
Different Utility Functions

• Example uk(rk)1-akrk
• With rk0 ? Karush-Kuhn-Tucker conditions
  (reverse water-filling)
• Better than utility-oblivious fair sharing

Equal-slope Pareto condition
uk
Vilfredo Pareto 1848-1923
rk
11
Packet Losses

• Myth Compressed streams are extremely
  vulnerable to losses. Even a single flipped bit
  might lead to a catastrophic decoding failure.
• Reality
• Packets are dropped, if CRC indicate bit error(s)
• Packet-level error control
• ARQ (TCP uses Go-back-N)
• FEC across packets with erasure decoding
• Hybrid ARQ Send parity packets instead of
  retransmissions
• ? Reliability can achieved at the expense of
  delay
• Compressed audio and video streams contain
  residual redundancy for robustness (graceful
  degradation)

12
FEC of Lost Packets

• Any k out of n packets allow reconstruction of k
  original information packets
• Not well suited to changing loss rates ? hybrid
  ARQ
• Increased delay, when packets are lost
• Extension to multiple description coding

x
E N C O D E R
D E C O D E R
x
x
x
k
k
x
n
k n
13
Packet Path Diversity

• Idea set up multiple connections along different
  paths
• Improved congestion resiliency
• lower combined latency
• better loss characteristics

D
Relay
1
Relay
2
cross traffic
cross traffic
Media traffic
S
14
Adaptive Playout with Two VoIP streams
15
Multiple Description Coding for VoIP
16
Reduced Delay w/ Packet Path Diversity
17
Flow Distribution for Minimum Congestion

• Fully connected network, link capacities decrease
  with distance
• Random cross traffic
• Streaming 100 kbps from Node 1 to Node 5
• Congestion is expected delay of packet

18
Increased Throughput with Multipath Routing
19
Suggestions for a New Internet

• QoS vs. best effort
• Share links based on flow utilities
• Rate allocation also provide admission control
• Error control
• Accommodate application-specific delay/loss
  tolerance
• Hybrid ARQ strong candidate better than Go-Back-N
• Multiple description coding?
• Flow-based multi-path routing
• Load-balancing reduces congestion/delay
• Send redundancy without increasing delay
• Elegant solution for hand-over problems
• Retire TCP error control, congestion control and
  flow control dont mix

20
The End