To design and develop a dynamic bandwidth management model

Download Report

Transcript To design and develop a dynamic bandwidth management model

Dynamic Bandwidth
Management
in QoS aware IP Networks
Committee:
Dr. Paul Farrell
Dr. Javed Khan
Dr. Hassan Peyravi (Chair)
Thesis
Yasir Drabu
Presentation Overview

INTRODUCTION



BACKGROUND WORK





Related Work
Proposed Model – Analysis and Algorithms
SIMULATION


Traffic Management - Active Queue Management (AQM),
Scheduling
Admission Control – Classification, Current Implementation and
limitations.
DYNAMIC BANDWIDTH MANAGEMENT (DBM)


Application, Current Network Problems, QOS, Traffic
Problem Definition and Research Goal.
Setup - Topology, traffic models, parameters and scenarios.
RESULTS
CONCULSION and FUTURE WORK
October 31, 2003
Dynamic Bandwidth Management
2
Applications – Changing Needs

Conventional Apps

Email, FTP, Telnet, etc
 Loss sensitive, High delay
tolerance, jitter insensitive.
 The IP Network was designed
for these.

New Applications

WWW started a new trend.
 Video, VoIP, Interactive/
Streaming Video, ecommerce, etc.
 Loss tolerant, delay sensitive,
jitter sensitive to varying
degrees.
 The IP Network was not
designed for these.
October 31, 2003
U
T
I
L
I
T
Y
U
T
I
L
I
T
Y
Bandwidth
Traditional Applications
Dynamic Bandwidth Management
Bandwidth
New Real-time
Applications
3
Current Networks - Issues

Inherent Problems
 Different

traffic requirement, similar treatment.
Signaling packets, Real-time packets, data packets,
individual packets within a flow, all treated same.
Ill behaved traffic hurts well behaved traffic.

Unresponsive UDP flows dominate TCP flows.
 Congestion

Control limited to end hosts.
TCP is predominant means of congestion control.
 Changing/
Upgrading Difficult.
Bottom line – Need Quality of service
October 31, 2003
Dynamic Bandwidth Management
4
Quality of Service – QoS

User QoS


Application QoS


Highly perceptional, hard to quantify.
Applications change so do
requirements.
Network QoS

Easy to quantify, well defined.
 All other QoS can be expressed in
these terms.
 Metrics well defined.





Availability
Delay
Delay Variation
Throughput (Bandwidth)
Packet Loss Rate.
October 31, 2003
Fig: Core Network QoS metrics
Dynamic Bandwidth Management
5
Network QoS – Approaches
Best Effort Enhancements (RED, WRED,
ECN)

Pros: Implemented over existing
infrastructure.
 Cons: No QoS guarantee

Integrated Services



Hybrid
Best
Effort
Hard Qos
Soft QoS
Cost
Differentiated Services



Pros: High level of QoS
Cons: Not Scalable, consistency issues,
implementation a big problem.
Complexity

Pros: Scalable, incremental implementation
Cons: QoS relative, unable to control flow
misbehavior within aggregate.
Constrain Based Routing


Pros: Scalable, better network utilization
Cons: Complexity (computational and
space), state information coherence, routing
stability.
October 31, 2003
Courtesy Cisco
Dynamic Bandwidth Management
6
Today’s IP QoS technology
Technology Description
Engineering
Aspect
RSVP
DS Byte
Out of Band Signaling
In Band Signaling
Signaling
CAR (committed
access rate)
Classification and policing
(application, protocol , DS Byte)
Policing &
Classification
RED, WRED
Weighted Random Early Detection
Service class enforcement
Congestion
Avoidance
WFQ, CBQ
Weighted Fair Queuing
Class-Based Queuing
Queuing Policies
Congestion
Management and BW
Allocation
MPLS
MPLS Diffserv
IP+ATM QoS integration
Leverage
Layer2
October 31, 2003
Dynamic Bandwidth Management
7
Today’s Internet Traffic

Conventional Traffic (Exponential, Smooth)



Exponential like Voice
Easier to analyze
Concrete parameters



arrival rate, queuing delay, etc.
Can be simulated using Poisson's Distribution
Actual Traffic (Self Similar, Bursty)



Heterogeneous mix of data, voice and multimedia application
Difficult to characterize. In some cases not possible to
characterize.
Effects of multiplexing




Makes the aggregate more self similar
Makes the traffic more exponential (contradicting)
Is there a factor, that can be used to decide the actual effect?
Can be simulated using Multiple Pareto distributions.
October 31, 2003
Dynamic Bandwidth Management
8
QoS – Bandwidth Allocation
Problem


Fixed Bandwidth allocated for QoS guarantee
Admission Control







Requires a priori information about traffic characteristics.
Traffic model does not accurately describe statistical behavior.
User defined parameters may not accurately represent the actual
traffic.
LDR traffic compounds the issue.
Some work in Call Admission Control – MBACs, little to no work
in Aggregate Flow control.
Weighted Scheduling – Weights associated based on
admission rates.
Bottom Line – Bandwidth Allocation is inefficient.
October 31, 2003
Dynamic Bandwidth Management
9
Problem Definition and Goal
“To design and develop a dynamic bandwidth management model
for efficient utilization of a shared link in a QoS aware IP
Network.”

Design Guidelines:
1.
Optimize bandwidth Utilization

2.
Lower Loss rate

3.
Non responsive UDP type flows can be controlled.
Coexistence with Best effort

5.
Minimizes bad drop/mark decisions.
Fairness

4.
Better Delay Vs Utilization trade off.
Prevent starvation of BE traffic
Scalable and Easily deployable

October 31, 2003
The model has to be scalable on a huge network and be incrementally
deployed.
Dynamic Bandwidth Management
10
Background Work

Three major components:



Active Queue Management (AQM)
 Random Early Detect
 Exponential Weighted Measured
Average (EWMA)
Admission Control – Arrivals
Scheduling – Service Allocation
Longer
Network Provisioning
Days
Hours
Performance Management
Minutes
QoS Re-routing
Bandwidth Reallocation
Seconds
Admission Control
AQM
Active Queue Management
Milliseconds
Scheduler
Microseconds
Admission
Control
October 31, 2003
Scheduling
Dynamic Bandwidth Management
Shorter
Time Scale
11
Typical QoS Aware Interface
Met
erin
g

Classifier
IN
Admission
Control
S
OUT
Usually a QoS Enabled Edge Router has –
 Classifier, Admission
controller, per class queue,
scheduler.
 Optional Metering unit.
October 31, 2003
Dynamic Bandwidth Management
12
Classification – Traffic Management
Active Queue
Management
Non-Reservation
Based
Proactive
FIFO
PQ
Reservation
Based
Router
Centric
FQ
Policing
Static
WFQ
WRR/CBQ
DWRR
Scheduling
Back
Pressure
Traffic
Management
Adaptive
Reactive
AWRR/ACBQ
ECN
Host
Centric
October 31, 2003
Dynamic Bandwidth Management
13
Classification – Admission Control

Extensive research
on Call Admission
Control.
Direct Probing
Host centric
for more
efficient bandwidth
allocation.

Packet Admission
Control, limited
work as compared
to CAC.
Early
Rejection
Packet Admission
Control
 MBAC
Slow Start
AAC
Router centric
CAR
Measurement
Based
Admission
Control
Call Admission Control
Simple Sum
Measured
Sum
Hoeffding
Bounds
Statistical
Parameter
Based
Non-statistical
Policy Based Admission Control
October 31, 2003
Dynamic Bandwidth Management
14
Dynamic Bandwidth Management

To over come the inefficient bandwidth allocation we
use dynamic approach.
Measure traffic conditions “online” and make admission
and allocation decisions.
Three approaches – closed loop, open loop and hybrid.

Closed loop




Queue Length
Closed Loop
Loss
Dynamic Bandwidth
Management
Open Loop

Delay
State information used as
feedback
Prediction based on Past
observations.
October 31, 2003
Dynamic Bandwidth Management
Rate
Open Loop
Hybrid
15
DBM – Proposed Architecture

On QoS Enabled Router Interface:




Introduce a controller
Feedback from Token Bucket - Starvation Rate tells us rate of packet arrival
Feedback from Queue – Average Queue Size tells us rate of packet
departure.
The Control has two components


Adaptive Admission Control (AAC)
Adaptive Class Based Queuing (ACBQ)
Adjusted Token Rates
r3
r2
r1
r0
B
B
B
B
3
2
1
0
Controller
w0
w1
w2 S
w3
Adjusted Scheduler Weights
Per Class Queue
October 31, 2003
Dynamic Bandwidth Management
16
The Controller Components
Dynamic Bandwidth Controller
Adaptive Admission Control

Adaptive Scheduler
Main Function:

 Monitors Arrival
Rate
 Decides Packet
admission

 Monitor
Queue State
 Decides BW allocation

Design Parameters:

Weight ( wi )
 Queue Thresholds (Thi )

Decision Parameter: ( si )
 Bucket
October 31, 2003
Design Parameters:
 Service
 Bucket
Size ( Bi )
 Token Rate (ri )
Main Function:
Decision Parameter:
 Average
Queue Size (ˆ i )
Starvation Rate
Dynamic Bandwidth Management
17
Adaptive Admission Control Analysis

Bucket Starvation Rate
 This indicates how
many tokens a flow
needs.
 Fuller buckets mean
lesser requirement
and vice versa.
 : Link Capacity
Startvation Rate :
si (t )  1  bi / Bi
Normalize Token Distributi on :
m
pi (t )  si (t ) /  s j (t ), 1  i  m
j 1
The Averaged Strarvation Rate :
pˆ i (t )  (1   ) pˆ i (t  1)  pi (t )
New Token Rate
bi : Tokens in the Bucket of i thClass ri  pˆ i (t ) 
m : Number of Classes
Bi : Bucket Size of i thClass
October 31, 2003
Dynamic Bandwidth Management
18
AAC: Algorithm
October 31, 2003
Dynamic Bandwidth Management
19
Adaptive Scheduler (ACBQ) Analysis

Each Queue has




TH – Upper Threshold
Average Queue Size ̂ i
For Lesser Delay lower
TH .
Average Queue Size
indicates w.r.t to TH tells
us how much service the
queue class requires.
October 31, 2003
Absolute Service Starvation Rate :
q (t )  ˆ (t ) / T
i
i
hi
Normalized Service Stravation Rate :
n
wi (t )  qi (t ) /  q j (t )
j 1
Average Service Rate :
i (t )  wi (t ) 
Dynamic Bandwidth Management
20
ACBQ – Algorithm
The parameters:
Wq  typically set to 0.002
typical _ transmission _ time  0.001
October 31, 2003
Dynamic Bandwidth Management
21
Simulation Setup
October 31, 2003
Dynamic Bandwidth Management
22
Simulation Scenarios
No. Scenario Name
Traffic
Load
1
Base Line
Exp
0.3-0.95 None
Static CBQ
2
Static Allocation
Exp
0.3-0.95 CAR
Static CBQ
3
Partial Adaptive
Exp
0.3-0.95 AAC
Static CBQ
4
Fully Adaptive
Exp
0.3-0.95 AAC
Adaptive CBQ
5
Base Line
Pareto
0.3-0.95 None
Static CBQ
6
Static Allocation
Pareto
0.3-0.95 Car
Static CBQ
7
Partial Adaptive
Pareto
0.3-0.95 AAC
Static CBQ
8
Fully Adaptive
Pareto
0.3-0.95 AAC
Adaptive CBQ
October 31, 2003
Dynamic Bandwidth Management
AC
Scheduling
23
Data Collections

Each simulation scenario was run
 3 times with different seeds
 For a duration of 70 seconds



Data was collected between 10-70 seconds,
assuming the simulator too 10 seconds to reach
steady state.
The instantaneous values per simulation
scenario were averaged over the duration of the
simulation.
Then again the averages were averaged for the
three runs.
October 31, 2003
Dynamic Bandwidth Management
24
Delay Performance – Exponential
October 31, 2003
Dynamic Bandwidth Management
25
Delay Performance – Pareto (LRD)
October 31, 2003
Dynamic Bandwidth Management
26
Jitter Performance - Exponential
October 31, 2003
Dynamic Bandwidth Management
27
Jitter Performance - Pareto
October 31, 2003
Dynamic Bandwidth Management
28
Loss Rate – Exp and LRD
October 31, 2003
Dynamic Bandwidth Management
29
Offered Load Vs Throughput
Average: 5% BW saving
At 80% Load – 14.5% BW saving
October 31, 2003
Average: 11% BW saving
At 80% Load – 30% BW saving
Dynamic Bandwidth Management
30
Simulation Results Summary

Higher Efficiency
 Lower
Packet Drop - Prevented bad admission
decisions.
 Increased Throughput


Bursty traffic showed better gain.
Tradeoffs for Higher Throughput
 Increased
Delay
 Increased Computation to O(N), Where N is the
number of QoS classes. N is always small.
October 31, 2003
Dynamic Bandwidth Management
31
Conclusion

CONTRIBUTION
 New
approach to Bandwidth Management
 Our approach performs better than commercial
implementing in terms of bandwidth utilization.

FUTURE WORK
 Define
an accurate relation between the QOS metrics
and Control Parameter.
 What is the best time scale of operation?
 How does it behave with RED and its Variants?
 End to End QOS is still not addressed.
October 31, 2003
Dynamic Bandwidth Management
32
Thank you.
October 31, 2003
Dynamic Bandwidth Management
33
View On Fundamental Limitations

Application Utility as a function of Network
Performance
 This

is undefined, and needs to be well defined.
QOS tries to provide better than BE
 How
about worse than BE, Scavenger service, “nice”
in UNIX process resource sharing.
 Elevated services, non-elevated services (Internet2).

Deployment is a bigger challenge that most
people think.
October 31, 2003
Dynamic Bandwidth Management
34
2 Ethernet and 8 Slip Interface Node Model
Opnet
R
ou
ter
N
od
e
M
od
el
IP Child Process
Model:
The child process
model, ip_output_iface
implements all the
Scheduling and AQM
algorithms like Class
Based Queuing, RED,
WRED, etc.
IP
Pro
ces
s
Mo
de
ip_dispatch Process
Model
Child Process
October 31, 2003
Dynamic Bandwidth Management
IP Node Model:
The router
implements the
complete networking
stack. We are
interested in the IP
node. We modify
this node to allow us
to probe some of the
statistics that we
collect in relation to
our proposed model.
We also modify the
underlying process
models by
implementing our
adaptive algorithms.
IP Process Model:
The IP process model
has several child
process models like
ip_icmp, ip_vpn etc.
Each child process
model implements a
specific feature of IP.
35