Transcript Campus QoS Cost Issues Analysis

Pricing QoS
in
Campus Networks
Marcos Pinto – CityTech (CUNY)
SITE Orlando, FL March 2006
Slide 1
Quality of Service (QoS)
Network Performance Guarantees To Applications
Specifically:
. peak and average bandwidth,
. delay,
. jitter,
. packet loss, and
. packet error rates.
Marcos Pinto – CityTech (CUNY)
Slide 2
Campus Network
Backbone that connects many backbones spanning several
buildings at a single location.
Campus
Headquarters
Internet
(LAN H)
Department X
Department Y
(LAN X)
(LAN Y)
Router/Gateway
Laboratory
(LAN L)
Marcos Pinto – CityTech (CUNY)
Slide 3
Campus QoS Effect
Best-Effort, Single-Class-of-Service Internets
Into
Systems with Levels of Preferred Service
To
Certain Classes of Users or Applications.
Marcos Pinto – CityTech (CUNY)
Slide 4
“On-Campus" QoS
versus
“Off-Campus" QoS
. Cost of Local vs. Wide-Area Bandwidth
. Ownership of Network Resources
Result:
Simpler/Less Costly Solutions for On-Campus QoS
Marcos Pinto – CityTech (CUNY)
Slide 5
QoS for IP Networks
Internet Engineering Task Force (IETF)
proposes three schemes:
. IntServ (Integrated Service Model)
. Diffserv (Differentiated Service Model)
. MPLS (Multi-Protocol Label Switching)
Marcos Pinto – CityTech (CUNY)
Slide 6
QoS for IP Networks
. IntServ
. Complex Implementation (tracking each flow)
. Not Scalable
. Diffserv
. Low-Cost (aggregates flows)
. Scalable
. MPLS
. Protocols Not Yet Standardized
. Uses ATM header (small) in the IP packet
Marcos Pinto – CityTech (CUNY)
Slide 7
DiffServ QoS for IP Networks
. SLA prior to the use of Diffserv
. Flows of Same DiffServ Treated Equally
. Popular Research Topic
. Cost-Effective for LANs
. Too Expensive for WANs
Marcos Pinto – CityTech (CUNY)
Slide 8
DiffServ QoS for IP Networks
Congestion is less likely at the edges of the
campus network

No need for complex (costier) QoS-aware edge
devices
Marcos Pinto – CityTech (CUNY)
Slide 9
Pricing for IP Networks
Current State-of-Affairs in Campus Networks:
Network costs on a single campus tend to be relatively
fixed.

Operational costs recurring (salaries/upgrades)
But
Cost does not change much due to variations in usage.
Marcos Pinto – CityTech (CUNY)
Slide 10
Pricing QoS for IP Networks
Service Level Agreement (SLA):
. Contract between service providers and users
(The service provider agrees to provide a service with certain quality parameters so long as the customer’s traffic
satisfies certain constraints.)
SLA includes:
. Guarantee of Quality of Service
. Pricing
Marcos Pinto – CityTech (CUNY)
Slide 11
Pricing for IP Networks
Three Pricing Methods:
. Flat Charge
. Quota/Allocation
. Charge based on Usage
Marcos Pinto – CityTech (CUNY)
Slide 12
Pricing for IP Networks
Flat Charge
. Fixed amount of money independently of traffic volume
. Simple and easy to implement
. Penalize ‘light’ users
. Low efficiency

Not a good solution for pricing the multi-service networks
Marcos Pinto – CityTech (CUNY)
Slide 13
Pricing for IP Networks
Quota/Allocation - Rationing
. Allocation of network resources is controlled primarily by FIFO
. Quota system remains unable to individualize allocation levels to
the applications
. Quota system will ultimately need user participation
Marcos Pinto – CityTech (CUNY)
Slide 14
Pricing for IP Networks
Usage-Based Charge
. Volumes of traffic and/or Time durations of the sessions
. Network Usages of Network Services.
It gives the aggressive users incentives to ‘behave’
themselves, which leads to alleviation of congestion.
Marcos Pinto – CityTech (CUNY)
Slide 15
Network Pricing Objectives
. Generate Capital Funds
. Recover Costs
. Control Demand
. Provide Budget Predictability
Marcos Pinto – CityTech (CUNY)
Slide 16
Collecting Money for Network Services
1. What will actually be counted and charged for?
(e.g., Access eligibility, Traffic level, etc)
2. Who
will be charged for: individual or group?
(e.g., User ID, MAC (Ethernet) address, IP address, Physical port ID, etc)
Marcos Pinto – CityTech (CUNY)
Slide 17
Pricing Method: Quota x Usage
In QoS-enabled IP Networks:
. Best-effort service first, and if needed, the more expensive premium
service.
. Not enough capacity (heavy users) or capacity wasted (light users).
. Users will only use premium services when network is congested.
. Quota portion (at subscription) might have to be included in Usagebased pricing
Marcos Pinto – CityTech (CUNY)
Slide 18
Usage-based Pricing Schemes
These schemes are of two kinds:
1. Service Class-based:
a. Paris Metro Pricing
b. Priority Pricing.
2. Auction-based:
a. Smart Market Mechanism.
Marcos Pinto – CityTech (CUNY)
Slide 19
Usage-based Pricing Schemes
Disadvantages:
. Requires extra network protocols and infrastructure
. Measurement & collection of the billing-related Data (time and volume)
have a large overhead.
. Cost of measuring the packets greater than the actual value of the
packets themselves.
Marcos Pinto – CityTech (CUNY)
Slide 20
Time-Volume Pricing Scheme
According to the SLA:
. Customer must send at no more than a maximum rate h (the a priori information)
. A connection sends data at a mean rate m (the a posteriori information) (m <= h )
. Then, the model defines:
. The charge c(m) is:
c(m) = a(m)T + b(m)V
V = TM
, where
V = volume of traffic carried (in cells or bytes),
T = connection duration (in seconds), and
M = measured mean rate of the user’s traffic.
a(m), b(m) = traffic coefficients
Marcos Pinto – CityTech (CUNY)
Slide 21
Time-Volume Pricing Scheme
The Basic Idea of the Model:
Given all ingredients for computing an effective bandwidth, namely:
. the traffic mix,
. the buffer size,
. the capacity, and
. the cell loss guarantee,
then the effective bandwidth α(m) for a new connection with a traffic
profile (i.e., peak rate), but whose mean rate is not known, can be
seen as a function in the mean rate m.
Marcos Pinto – CityTech (CUNY)
Slide 22
Time-Volume Pricing Scheme
The effective bandwidth is plotted against the mean rate M for a fixed peak rate h.
The user is free to
choose any tangent
to this curve, and
is then charged
a(m) per unit time
and b(m) per unit
volume.
Marcos Pinto – CityTech (CUNY)
Slide 23
Time-Volume Pricing Scheme
. The effective bandwidth α(m) is a concave function
. It can be approximated by a number of straight lines of the form:
a + b·m = a + b·V/T
. If the user has the option to choose one of these as tariff a·T + b·V ,
then he will choose the (a,b) minimising a + b·m
. In this way, he chooses a tariff approximating α(m) and implicitly
provides an estimate of m to the network.
Marcos Pinto – CityTech (CUNY)
Slide 24
Simulation Using OPNET
. Two connected campus networks exchanging packets
. Routers as the agents to collect T and V
. Several runs with different (a,b) that minimize charges c
Charging
Agent
Charging
Agent
Charging
Agent
Src 1
Dest 1
Dest 2
Src 2
Ingress
Charging
Agent
Egress
Dest 3
Src 3
Egress
Diffserv Domain
Diffserv 1
Marcos Pinto – CityTech (CUNY)
Diffserv 2
Slide 25
Simulation Results Using OPNET
High-Priority Traffic
Marcos Pinto – CityTech (CUNY)
Low-Priority Traffic
Slide 26
Pricing QoS for IP Networks
Summary:
. Group of services to group of users with different service requirements.
. Networks are increasing in cost to universities
. Pricing becomes an important issue
. Charging, a big cultural change, and new policies need to be in place.
. A mere summary report of user’s usage could be suffice to deter abuses
Marcos Pinto – CityTech (CUNY)
Slide 27