halina tarasiuk, robert janowski and wojciech burakowski warsaw university of technology, poland...
TRANSCRIPT
Halina Tarasiuk, Robert Janowski and Wojciech Burakowski
Warsaw University of Technology, Poland
Admissible Admissible TTraffic raffic LLoad of oad of RReal eal TTime ime CClass of lass of SService for ervice for
IInter-domain nter-domain PPeerseers
Contents
• Classes of service concept as an approach for providing strict QoS guarantees at the network level– Experiences from AQUILA project (5FR)
– EuQoS project (6FR – in progress)
• RT service at an inter-domain peer
• CAC for RT service– Algorithm
– Numerical results
• Summary
Classes of service concept as an approach for providing strict QoS guarantees at the network level– Experiences from AQUILA project (5FR)– EuQoS project (6FR – in progress)
QoS at different levels
user user
codec codec
Additional mechanisms
Additional mechanisms (e.g. playback buffer)
Network interface
Network interface
Application level
Application level
ITU G.1010
ITU Y.1541
Subjective assessment
Network level
Network level
User level User level
To guarantee packet losses, packet delays
Subjective assesment
Classes of Service
Class of service concept
• A „service class” represents a set of traffic that requires specific delay, loss and jitter characteristics from the network for which a consistent and defined per hop-behaviour applies
• A service class pertains to applications with similar characteristics and performance requirements
Discussed Classes of services (IETF proposal)
Tolerance To Tolerance To Inter-Provider
Service Class (Aggregate)
Loss Delay Jitter PHB
End-To-End Service Class Loss Delay Jitter
DSCP Name
DSCP Value
Ctrl Low Low Yes CS Network Control
Low Low Yes CS7 111000
Telephony VLow Vlow VLow EF 101110 Signalling Low Low Yes CS5 101000
MM Conferencing
L-M Vlow Low AF4x 100xx0*
RT Interactive
Low Vlow Low CS4 100000 Real Time VLow VLow VLow EF
Broadcast Video
VLow Med Low CS3 011000
MM Streaming
L-M Med Yes AF3x 011xx0*
Low Latency Data
Low L-M Yes AF2x 010xx0*
OAM Low Med Yes CS2 010000
None Real Time
Low LIM Yes AF
High ThruPut Data
Low M-H Yes AF1x 001xx0*
Best Effort NS NS NS DF Standard NS NS NS DF 000000
• End-to-end – related to applications (visible by users)
• Aggregated in some network parts (maintained by the network)
Definition of a service class
1. QoS objectives: values of packet losses, delays...
2. Types of connections: p2p
3 Traffic descriptors: single-, double token bucket, more advanced
A. Provisioning of resources: static, dynamic
B. CAC: based on declarations, based on measurements
C. Tuning mechanisms at the packet level (PHB: classifiers, scheduling, marking, active quieueing..)
Experiences in implementing CoSs -
AQUILA network (2000-2003)
(IST-1999-10077)
Adaptive Resource Control for QoSAdaptive Resource Control for QoSUsing an IP-based Layered ArchitectureUsing an IP-based Layered Architecture
AQUILA
Admission ControlAdmission
Control AgentAdmission
Control AgentEnd-userApplication
Toolkit
Resource Control Layer
Core Router
Core Router
Core Router
Access Network
Access Network
Edge Router
Edge Router
Set
tin
g
Resource Control
Resource Control Agent
Set
tin
g
Consideration of Network LoadMonitoring
ProbingResults
AQUILA Architecture
resources resources
QoS Request
QoS Request
QoS Request
QoS in core networks – IP prototype solutions: AQUILA
Network service Traffic type
Characteristic examples
Application example
Premium CBR Constant
Small packets low loss, low dela
SIP VoIP
Premium VBR Variable Large packets low loss, low delay
SIP Video
Premium MM Adaptive required throughput
File transfer (FTP)
Premium MC Very short bursts
very low delay & loss
online games
Standard Best effort classical the rest
Goal: only a few few network services to allow clear service differentiation
Tested CAC algorithm for PCBR service - RT service
New flow is admitted if:
(1)
Where N1 denotes the number of connections in progress and parameter (<1) specifies the admissible load of capacity allocated to the PCBR. The value of can be calculated from the analysis of M/D/1/B system taking into account the target packet loss ratio and the buffer size [2].
(2)
Where Buffer denotes buffer size in packets and Ploss target packet
loss ratio.
1
1
1
CPBRPBRN
iinew
)ln(2
2
lossPBuffer
Buffer
Overall Topology for Trial in AQUILA
Implementing CoSs in EuQoS system (2004-2006)
End-to-end Quality of Service support over heterogeneous
networks
Some of the problems to be solved
• Scalable architecture
• Signalling system
• Providing QoS at the packet level
• To cope with network heterogeneity
• Etc.
USER 1 USER 2
EQ-SDP
EQ-SIPSignaling
EQ-SIPSignaling
Network technology Independent sub-layer
Network technology dependent sub-layer
EQ-SDP in
End-to-end QoS EQ-SIP signaling
AccessNetwork
1
QoSDomain
i
AccessNetwork
2
QoSDomain
k
QoSDomain
j
RA1 RAkRAjRAiRAn
RM1 RMi RMj RMk RM2n
Virtual Network Layer
Application Application
EuQoS Architecture: Physical View
EQ-pathEQ-path
EQ-SDP
EQ-ETP
Protocols
EQ-ETP
Protocols
EQ-NSIS EQ-NSISEQ-NSIS EQ-NSIS
EQ-SIPproxy
EQ-SIPproxy
EuQoS system
End-to-end CoS path
QoS domain path QoS domain path QoS domain path QoS inter-domain path
QoS inter-domain path
QoS signalling – EQ-SSN
QoS routing – EQ-BGP (path)
RM
QoS Request
QoS Request QoS Request
QoS Request QoS Request
RA
QoS Request RM
RM
CoS CoS CoS CoS CoS
Borders for Classes of service
Intra- and inter-domain Classes of service
Inter-domain service AS1-AS2
AS1
output port
Egress BR
AS2
Intra-domain service in AS2
input ports Router input port
Ingress BR
Ingress BR
Ingress BR
Intra-domain service in AS1
output port
AC AC
AC
AC
AC: admission control
Classes of Service in EuQoS
QoS Objectives End-to-end Class of
Service IPLR Mean IPTD
IPDV Type of
connections Traffic descriptors
DSCP code
Telephony 10-3 100 ms 50 ms
P2P Peak bit rate, single token bucket
101110
RT Interactive 10-3 100 ms 50 ms
P2P Peak bit rate, single token bucket
100000
MM Streaming 10-3 1 s U P2P Peak bit rate, single token
bucket4
011xx0*
High Throughput Data
10-3 1 s U P2P Peak bit rate, single token
bucket4
001xx0*
Standard U U U 000000
4 We assume that for NRT traffic we will use single token bucket descriptor with given both bit rate and bit rate tolerance, depending on the application and network technology.
EuQoS Applications (Phase 1) Medigraf End-To-End
Classes of Service
VoIP VTC VoD VTC
Collaboration
data transfer
Chat
Telephony X RT
Interactive X X
MM Streaming
X
High ThruPut
Data
X
Standard X
Plan for developping CoSs in EuQoS
Access 1 (LAN)
Access 2 (UMTS)
Core
Telephony
RT Interact.
MM Stream.
High Thr. D.
Standard
RT
NRT
BE
Over-provisioned
RT
NRT
BE
User request for basic end-to-end
CoS
Aggregated CoSs
implemented on inter-
domain link
Aggregated CoSs
implemented on inter-
domain link
Aggregated CoSs
implemented in core domain
Aggregated CoSs implemented inside domain
mapping mapping mapping mapping mapping
RT
NRT
BE
RT
NRT
BE
Aggregated CoSs implemented inside domain
Access networks: LAN/Ethernet, xDSL, WiFi, UMTS
IP core: Geant
EuQoS Test Network
Applications vs. Classes of Service End-to-end
CoSs
Telephony
High throughput data
MM
streaming
RT interactive
EUQOS Applications
VoIP
Data Transfer
VoD
VTC
EUQOS Applications
VoIP
Data Transfer
VoD
VTC
End-to-end CoSs
Telephony
High throughput data
MM
streaming
RT interactive
Real Time Real Time RT
CoS on Access Network
CoS on Acess Network
CoS in Inter-domain Link
RT
NRT1
NRT3
NRT2
NRT1
NRT3
RT
NRT1
NRT3
NRT2
EQ-PATH
Network Control
CTRL CTRL CTRL Network Control
• RT service at an inter-domain peer• CAC for RT service
– Algorithm
– Numerical results
RT Class of Service
• End-to-end Classes of Service– Telephony for VoIP – short packets (60 bytes)
– RT Ineractive for VTC – long packets (1500 bytes)
• QoS metrics– IPLR – 10^-3
– Mean IPTD – 100 ms
– IPDV – 50 ms
• Traffic description: single token bucket (PBR, PBRT)
• Policing strategy– Policing in access network only (entry point); to police (PBR, PBRT)
– We have to define in each access network policy point (node)
Approach 1: not distinguishing between e2e CoSs
• CAC algorithm
• It does not take into account an impact of packet sizes on IPLR
RTRT
N
iinew CPBRPBR
1
1
)ln(22
IPLRBuffBuff
RT
RTRT
Approach 2:
Studied system for RT service for inter-domain pear
Assumptions:
- the input traffic of both end-to-end CoSs is Poisson process.
- the packet sizes are constant equal to ‘d1’ and ‘d2’, respectively for telephony and video conference CoSs and their ratio (d2/d1) is an integer denoted by ‘d’.
- the packets of these CoSs enter the same finite buffer (with buffer size – Buffer counted in packets).
Analysis (1)
Time
Packet Arrivals
Packet Departures
Q(n-1) Q(n)
Q(n+2) Q(n+3)
Q(n+1) Q(n+4) Embedded instants of inspections of system state
empty was_system_ 21
_servedpkt_type2_ 1
21
11)(
_servedpkt_type1_ 211)(
)1(
ifAA
ifd
jA
d
iAnQ
ifAAnQ
nQ
Where:
-Q(n) denotes the system state at the end of n-th embedded time instant
- A1, A2 random variable describing the number of type 1 (respectively type 2) packet arrivals during one slot,
-Ratio of packet sizes is denoted as ‘d’ (d2/d1 = d)
Figure 8. Time evolution of the system state
Analysis (2)
the load () and the arrival intensities (1,
2) are related by:
1=1d1=1 (since d1=1);
2=2d2=2d; =1+2
1=w1 ; 2= w2
After some algebra
Eq.9
)1)(()1)((
)1)(()1)(()1)((
2221
21
21211
21
1
1211
21
22221
21
11211 )1())(1()(
zddzdd
zddzddzdd
e- ez
zee ezQ
Analysis (3)
x
zCxQob )
1(}{Pr
00
Assuming that the tail probabilities of the queue size distribution function are well approximated by the dominant pole of Q(z), they
can be written as
Further, assuming that the asymptotic constant Co equals 1, the buffer overflow probability can be expressed as
1
002 00 )
1(
1
1)
1(
}1{Pr
Buffer
Buffern
n
zzzC
BufferQobPloss
Eq.12
Analysis (4)
we can determine the value of the required decay rate parameter 1/z0. This decaying rate ensures that the buffer overflow probability will be below target Ploss value.
0
)1)((
1
)1)((
1
10
021
2
2
02
1
2
zwdw
dw
dw
zd
ww
dw
ew
ew
wz
Steps to calculate the admissible load when all the input parameters (Buffer, Ploss, d1, d2, percentage contribution of different types of traffic - w1, w2) are given:
1. Given Ploss and Buffer, determine the parameter z0 (Eq.12)
2. Create the equation (14) taking into account the number of traffic types, their characteristics (intensity, packet sizes) and the assumed input model (Poisson).
3. Solve the equation (14) with respect to which is the total admissible load.
4. Calculate the admissible load of each traffic class based on the information about percentage contribution of different traffic classes - w1, w2 (9), i.e. 1=w1, 2=w2
Eq.14
Numerical results (1)
Figure 9. Total admissible load vs. packet size ratio of two end-to-end CoSs; target Ploss=10-3
Figure 10. Packet loss ratio vs. packet size ratio of two end-to-end CoSs; target Ploss=10-3
0.00001
0.0001
0.001
0.01
0.1
1
1 3 5 7 9 11 13 15d
Plo
ss
Target Ploss
I-10% / II-90%
I-50% / II-50%
I-90% / II-10%
0
0,3
0,6
0,9
1 3 5 7 9 11 13 15d
tota
l ad
mis
sib
le l
oad
I-10%/ II-90%
I-50%/ II-50%
I-90%/ II-10%
7
Numerical results (2)
0.00001
0.0001
0.001
0.01
0.1
1
1 3 5 7 9 11 13 15d
Plo
ss
Target Ploss
I-10% / II-90%
I-50% / II-50%
I-90% / II-10%
Figure 11. Packet loss ratio vs. packet size ratio of two end-to-end CoSs; target Ploss=10-2
Figure 12. Packet loss ratio vs. packet size ratio of two end-to-end CoSs; target Ploss=10-4
0.00001
0.0001
0.001
0.01
0.1
1
1 3 5 7 9 11 13 15d
Plo
ss
Target Ploss
I-10% / II-90%
I-50% / II-50%
I-90% / II-10%
Summary
• QoS guarantees at the network layer we can assure by providing classes of service
• RT service for inter-domain peers requires adequate CAC algorithm
• The proposed algorithm works correctly and takes into account differences in packet sizes
• The algorithm will be implemented in EuQoS system and tested
• „Admissible traffic load of real time class of service for inter-domain peers” in Proc. of ICAS/ICNS 2005, 23-28 October 2005, Papeete, Tahiti, French Polynesia, published by IEEE Computer Society, 2005.
• The full text paper can be found at the homepage of TNT Group http://tnt.tele.pw.edu.pl/include/members/Artikuly/Admissible.pdf