Resource Management in IP Telephony Networks

1
Resource Management in IP Telephony Networks

• Matthew Caesar, Dipak Ghosal, Randy H. Katz
• mccaesar, randy_at_cs.berkeley.edu
• ghosal_at_cs.ucdavis.edu

2
Motivation

• What is IP Telephony?
• Packetized voice over IP
• PSTN access through Internet Telephony Gateway
  (ITG)
• Benefits
• Improved network utilization
• Next generation services (POTS ? PANS)
• Growth
• Revenues 1.7 billion in 2001, 6 of
  international traffic was over IP, growing Frost
  2002 Telegeography 2002
• Standardized, deployed protocols (TRIP, SIP,
  H.323)
• ? Requires scalable architecture to limit
  congestion.

3
Goals

• High quality, economically efficient telephony
  over the Internet.
• Low blocking probability
• Provide preferential treatment, high QoS
• Questions
• How to perform call admission control?
• How best to route calls through converged network?

4
Approach

• Mechanisms
• Congestion sensitive call admission control
• ITG selection
• Techniques
• Awareness of ITG congestion
• Path quality between important points in network

Distance




ITG Utilization
5
Overview

• IP Telephony Networks
• Pricing-based Admission Control
• Redirection Techniques
• Experimental Design
• Results
• Future Work

6
System Architecture
ITG
ITG
LS
LS
ITG
LS
ITG
LS
ITG
LS
ITG
LS
Admin. Domain (AD)
Example Call Setup
Internet
Example Advertisement
Gateway (ITG)
ITG
IP Terminal
Example Call Session
Location Server (LS)
LS
7
Scope of Study

• All calls are net-to-phone
• ADs cooperate to provide service.
• Use IETFs TRIP architecture to support
  interoperability.
• Disregard degradation in access network.
• Prices determined at start of call.
• ITGs offer equal PSTN reachability.

8
Pricing

• PSTN
• distance pricing
• time of day pricing
• IP Telephony
• richer user interface
• allows for more dynamic pricing schemes
• Baseline Flat-rate Admission Control (FAC)

9
Congestion Sensitive Call Admission Control (CAC)

• Goal prevent system overload and generate
  revenue
• Price of call
• function of number of voice ports in use
• rises when highly utilized
• More dynamic than PSTN

10
Price-Congestion Function

• Used M/M/m/m (m-server loss system)
• responsive server
• loss system
• discouraged arrivals
• Found price-congestion function that maximized
  revenue with respect to a

11
Congestion Pricing Analysis

• Exponential function generates most revenue
• Stepwise linear function almost as good
• Maximum system price charged early
• Approximation to function minimizes price
  fluctuations

12
Redirection

• Problem finding the best ITG
• Approach tradeoffs between quality and load
• Method LS maintains
• Average measured path quality
• Number voice ports in use
• Algorithms
• Random Redirection (RR) (baseline)
• QoS Sensitive Redirection (QR)
• Congestion Sensitive Redirection (CR)
• Hybrid Scheme (CQR)

13
Redirection Schemes

• QoS Sensitive Redirection (QR)
• Different paths provide different service
• Technique
• Use RTCP RRs to monitor path congestion
• Route over best paths
• Congestion Sensitive Redirection (CR)
• Unbalanced load causes call blocks
• Technique
• Use TRIP advertisements to estimate ITG
  utilization
• Route to least utilized ITG

14
Hybrid Redirection (CQR)

• Choosing nearby ITG improves call quality, but
  can unbalance load.
• Algorithm
• Compute Rdm ? Mi(1-?)Qi
• Mi is utilization, Qi is loss rate
• Select randomly from k ITGs with lowest Rdm
• Tradeoffs
• Use ? to trade off call quality and load balance
• Use k to vary flash crowd protection
• Price Sensitive CQR (PCQR)
• Decrease ? for higher bids

15
Overview

• IP Telephony Networks
• Pricing-based Admission Control
• Redirection Techniques
• Experimental Design
• Results
• Future Work

16
Experimental Method

• Modified ns-2
• Ran for 1.5 simulated hours
• Eliminated first half-hour
• User Model
• Bid uniformly distributed
• Voice traffic on-off Markov process
• Self-similar cross-traffic
• Data points stable across several time scales

17
Evaluation Metrics

• Blocking Probability
• Average call QoS
• Used Mean Opinion Score (MOS) based on RTP loss
  rate
• Economic efficiency
• Ratio of service tier to QoS achieved

18
Admission Control Blocking Probability

• Flat pricing unnecessarily blocks many callers
• Congestion pricing changes system price
  dynamically with load

19
Redirection Blocking Probability

• Congestion sensitivity decreases blocking
  probability
• Small k ? few blocked calls
• Congestion Sensitive Redirection (CR) improves
  balance over Random Redirection (RR)

20
Redirection Background Traffic Effects

• QoS sensitivity minimizes effects of cross
  traffic
• Small amount of sensitivity vastly improves call
  quality

21
Summary

• Admission Control Schemes
• Congestion sensitive pricing decreases
  unnecessary call blocking, increases revenue, and
  improves economic efficiency
• Derived exponential price-congestion function
  that maximizes revenue
• Redirection Schemes
• Hybrid scheme achieves best of both worlds
• Price sensitivity improves economic efficiency

22
Future Work

• Realistic workload
• Improve user model
• Develop price-congestion function for real users
• Study flash-crowd effects
• ITG Placement
• Competitive Network

23
Resource Management in IP Telephony Networks

• Matthew Caesar, Dipak Ghosal, Randy H. Katz
• mccaesar, randy_at_cs.berkeley.edu
• ghosal_at_cs.ucdavis.edu