Multicast for Video Streaming
Download
Report
Transcript Multicast for Video Streaming
Wireless Networking
EE290T Spring 2002
Puneet Mehra
[email protected]
Topics
Supporting IP QoS in GPRS
QoS Differentiation in 802.11
802.11 and Bluetooth Coexistence
Bluetooth
Supporting IP QoS in the General
Packet Radio Service
GPRS – enhancement for GSM infrastructure to
support packet-switched service
Limitations in architecture:
Can only differentiate QoS on basis of IP address of
mobile station (MS) not on per-flow basis
GPRS core uses IP tunnels which makes
implementation of IP QoS difficult
Proposed Solutions
IntServ approach
DiffServ approach
GPRS architecture
GSNs – have GPRScompliant protocol stack.
QoS profile assigned to
every MS, but…
Supporting GSNs attach to
MS, Gateways attach to Net
No QoS in the network core
-> possible congestion
IP tunnels used between
GGSN and SGSN
So RSVP/Diffserv TOS bit
unavailable to intermediate
nodes
IntServ Approach to QoS
Establishing QoS across Core
Mapping RSVP QoS to GPRS QoS
Uses RSVP tunneling. Original
messages pass through, but then
additional state set up as needed.
GGSN coordinates all reservations
since it sees non-encapsulated
packets.
Use either UpdatePDPContextRequest
& ChangePDPContextRequest
messages, as well as
ModifyPDPContextRequest messages.
Requires significant changes to
GGSN, but other nodes just need
RSVP functionality
DiffServ Approach to QoS
• GGSN assigns incoming traffic to a specific PHB (figure 6)
• To provide QoS over MS <-> SGSN link, each MS has multiple IP’s.
• Each IP has own GPRS QoS and gets mapped to a given PHB class (can be done at
connect time or on demand).
• Requires significant changes to all components.
Simulation Environment
Random handoffs w/ A1
getting most traffic
Fast-moving and Slowmoving MS users modeled
Traffic reflected occasional
“rush hour” frequency
300,400 & 500 MSs
simulated for 4 hour periods
Results
Low Percentage of failed reservations
Low signaling overhead due to addition of
RSVP signaling
With 500 MSes, only 3.6% failed reservations
RSVP signaling was <2.5% of total traffic
Overall Good scalability due to RSVP
aggregation
Get even better performance if modify the RSVP
refresh interval
Evaluation of Quality of Service Schemes
for IEEE 802.11 Wireless LANS
802.11 has 2 different MAC schemes
Distributed Coordinator Function (DCF)
Point Coordinator Function (PCF)
4 Schemes Tested for Differentiation
PCF mode
Distributed Fair Scheduling
Blackburst
Enhanced DCF
802.11 Distributed MAC scheme
Carrier Sense Multiple Access with Collision
Avoidance (CSMA/CA) algorithm.
The Steps:
1.
2.
3.
4.
5.
First Sense the Medium.
If Idle for DIFS time period, send frame.
Else - do exponential random backoff involving
multiple of minimum contention window (CW)
Each time medium is idle for DIFS, window—
If(window == 0) transmit frame
Differentiation Methods
802.11e – Enhanced DCF
Different minimum contention window
Different interframe spaces
Higher priority has smaller window
Use Arbitration IFS – some multiple of DIFS time period
Packet Bursting – station can send multiple frames, for
certain time limit, after gaining control of medium
PCF
Centralized, polling-based mechanism involving the base
station.
Time consists of Contention Free Periods, when only polled
station access medium.
Differentiation Methods Cont.
Distributed Fair Scheduling (DFS)
Backoff interval dependent on weight of sending
station.
Blackburst
High priority stations try to access medium at
constant intervals.
Enter a blackburst contention period, where a
station jams the channel for time proportional to
how long it has been waiting.
Synchronization between high-priority flows leads
to little wasted bandwidth due to contention
Simulation Results
Simulations carried out in ns-2 with background cross
traffic
EDCF and blackburst provided best service to highpriority flows, especially with high loads, but starved
best-effort
Blackburst had best medium utilization
PCF performed worst, and EDCF is, distributed, and
offers better performance
DFS offered better service differentiation while
avoiding starving low-priority flows when network
load is high
Differentiation mechanisms for
IEEE 802.11
DCF Details
Hidden Node Problem
Solution – optional RTS/CTS scheme w/
fragmentation_threshold
Network Allocation Vector (NAV) used to do
virtual carrier sensing – get transmission
duration from RTS/CTS frame info
Different Inter Frame Spacing (IFS)
MAC ACK packets use Short IFS (SIFS) instead
of DIFS
QoS Differentiation in DCF
Backoff increase function
Different DIFS
Each priority level has a different backoff
increment function
Each priority has a different DIFS
Maximum frame length
Each priority has a different maximum
frame that can be transmitted at once
Backoff Increase Function
Original: backoff_time = Floor[22+i x rand()] x
slot_time
Modification: backoff_time = PJ2+i where PJ is the
priority factor. Larger value leads to longer
delay/lower throughput
Results
Provides differentiation for UDP, but large ratios lead to
instability
No effect for TCP. Assume that AP is responsible for sending
TCP-ACKs -> since senders ended up waiting for ACK from
AP and there was no contention for RTS messages
DIFS differentiation
Stations with higher priority have
smaller DIFS interval
Results
Works well for UDP flows
AP priority determines effect on TCP
differentiation (since it sends ACKs)
Can give UDP priority over TCP. How? By
changing priority of AP.
Maximum Frame Length (MFL)
Priority due to size of maximum
transmittable data unit
Results
Throughput proportional to MFL
Ratios don’t affect system stability
Can prioritize TCP or UDP traffic
Results of Channel Errors
All Approachs
Backoff Time Approach
Channel errors lower data rate
Prioritization dependent on channel (Bad!)
Maximum Frame Length
During channel errors, large packets more
likely to be corrupted -> smaller
differentiation
Wi-Fi (802.11b) and Bluetooth:
Enabling Coexistance
Bluetooth & WiFi Basics
Bluetooth - short range cable replacement tech. 1 Mb/s data
rate
WiFi - wireless LAN tech operating at 11Mb/s (actually up to
22Mb/s now)
Both Operate in 2.4 GHz Range
Bluetooth (uses FHSS) – transmit high energy in narrow
band for short time
WiFi (Uses DSSS) – wider bandwidth with less energy
Sharing spectrum -> interference
Interference Overview
Noise at Receiver
Types of Noise
In-band noise: noise in frequencies used (harder
to filter)
Out-of-band noise
White (Gaussian) – evenly distributed across band
Colored – specific behavior in time/frequency
To coexist:
Receivers must deal with in-band colored noise
but designed assuming only white noise
Interference Experiments
Experimental Setup
Used laptop w/ Wi-Fi and
bluetooth cards
Results
Wi-Fi stations less than 57m from AP suffered more
than 25% degradation in
presence of cubicle
environment
More Results
Bluetooth Throughput reduction due to Wi-Fi interference
Interference-Reduction
Techniques
Regulatory and standards
Usage and Practice
Eg: Allow bluetooth to only hop over certain range
Limit simultaneous usage to avoid interference
Technical Approaches
Limit bluetooth power for short-range devices
Use other frequencies (5 GHz – HiperLan and 802.11a)
Much more RF power required
Shorter Range
Appears to be an open research area
Bluetooth: An Enabler for
Personal Area Networking
Personal Area Network (PAN)
Electronic devices seamlessly interconnected to
share info (perhaps even constantly online)
Characteristics
Distributed Operation
Dynamic network topology (assume mobile nodes)
Fluctuating Link Capacity
Low Power Devices
Bluetooth’s role in PAN
Piconets
Adhoc networks formed
by nodes
Master/Slave semantics
with polling of data
Scatternet
Interconnection of
piconets.
Nodes may be in several
piconets at once, serving
as gateways
Routing Issues
Packet Forwarding in
Bluetooth
Bluetooth Network
Encapsulation Protocol
(BNEP) – ethernet-like
interface for IP
Scatternet forwarding –
use BNEP broadcast
messages and ad-hoc
routing techniques
Scheduling Issues
Intrapiconet Scheduling (IRPS)
Schedule for polling slaves in piconet
Interpiconet scheduling (IPS)
Scheduling a node’s time between multiple
piconets.
Main challenge: make sure that node is
available in piconet when master wants to
communicate
IPS Framework
Rendez-vous Point Algorithms
Proposed for IPS
Main Issues:
How to decide on the RP, and how
strict is the commitment
How much data to exchange during
RP
RP timing
nodes communicate when
slave/master will meet (in time) to
exchange data
can be periodic or pseudo random
Window exchange
Static or dynamic
References
“Supporting IP QoS in the General Packet Radio
Service”. G. Priggouris et Al. IEEE Network 2000.
“Evaluation of Quality of Service Schemes for IEEE
802.11 Wireless LANs”. Anders Lindgren et Al. IEEE
LCN 2001.
“Differentiation mechanisms for IEEE 802.11”. Imad
Aad and Claude Castelluccia. IEEE Infocom 2001.
“Wi-Fi (802.11b) and Bluetooth: Enabling
Coexistence”. Jim Lansford et Al. IEEE Network 2001.
“Bluetooth: An Enabler for Personal Area
Networking”. Per Johansson et Al. IEEE Network
2001.