ITU-T Kaleidoscope Conference Innovations in NGN TWO BUFFER

Geneva, 12-13 May 2008

ITU-T Kaleidoscope ConferenceInnovations in NGN

Gang QinThe University of Tokyo

[email protected]

TWO BUFFER MODEL-BASED QoSESTIMATION METHOD FOR 3G WIRELESS IP NETWORKS IN

BULLET TRAINS

Geneva, 12-13 May 2008 First ITU-T Kaleidoscope Conference – Innovations in NGN 2

Outline

Summary

The proposed measurement

The QoS estimation method

The experiment and result

Conclusions and Future Works


Summary

We are aiming at improving transport layer protocols over 3G wireless IP networks in high-speed mobile environment.

It is necessary to make a model of high-speed mobile communication environment on the simulator.

To do this, measurement of the communication quality and the raw packet transmission characteristics of the 3G is necessary

We will introduce a two buffer model-based QoSestimation method used to measure the raw packet transmission characteristics of CDMA2000 1xEV-DO in bullet trains.


Definition of communication quality and measurement(ITU-T Recommendations)

・IP packet transfer delay (IPTD) ・IP packet delay variation (IPDV) ・IP packet loss ratio (IPLR) ・IP packet error ratio (IPER)

• The measurement points (MP) are fixed and the path is invariable• Clock synchronization of two measurement points is necessary

• One-way measurement• The probe packet will be

• UDP–echo based• Time-stamped at injection

and extraction devices

Probe Packet

MP A MP B

Results


Issues of communication qualities in bullet trains1. The mobile station (MS) keep moving the entire time and

the communication channel characteristics differ according to the geographical location of the MS.• we need to define the invariant communication qualities of this

environment and find a method for the measure of communication performance.

2. In order to get the raw packet transmission characteristics of 3G wireless IP networks, we need to avoid the influence of possible packet buffers and transport layer protocols like W-TCP. • We can only observe the communication performance over IP layer.• We avoid measuring the network performance with tools using TCP.

3. The one-way measurement need to synchronize the clock of the two measurement points.• There are many tunnels along the route, so it is very difficult to realize

the clock synchronization by using GPS.


Proposed measurement

Define Route Characteristics in Bullet Trains as a set of the following measurements

RPB = {< Delayi, Jitteri, Packet_Lossi, Availabilityi>|i ∈ Locations}

Locations are sampled in a specified time-interval between the start station and the terminal station.Each measurement varies geographically.We want to get statistical values from RPB

For #1 issue


Proposed measurement

A proposed measurement from the view of users

Traditional two-way measurement (RTT)

ICMP-echo packets in a specified time-intervalThe size of probe packet is 84 bytes.The measurement interval is set to 1 secondMeasure from the starting station to the terminal station

Two buffer model-based QoS estimation method is used to measure the raw packet transmission characteristics

For #2 issue

For #3 issue

For #2 issue


How to measure:Network configurations of our experiment

Probe Packet

IP Network of au

PDSN1xEV-DO1xEV-DO The Internet

U. of Tokyo Network

IP Address given by au

Probing CEEcho CE

ACK Packet


What to Measure:The data of our experiment

We measure following data at each location segment i

RTT (Delayi, Jitteri)Packet Loss Ratio (Packet_Lossi )Probing CE Status (PCE_Statusi )

Experiment Target LineTokaido Shinkansen (Tokyo－Shin Osaka)「Nozomi」

No throughput measurement at this time


RTT data

RTT (ms) Tokyo-Shin Osaka

Shin Osaka-Tokyo Static

Min 140 132 131

Max 26199 24503 948

Average (mean) 661 575 189

Median 400 404 108

Standard deviation 1402 1088 42.6

Packet Loss Ratio 8.5% 7.9% 0%


Cumulative distribution of RTT

00.1

0.20.30.4

0.50.60.70.8

0.91

100 1000 10000 100000RTT(ms)

Tokyo-Shin Osaka Shin Osaka-Tokyo staticProbability


Cumulative distribution of packet loss length for Tokyo-Shin Osaka

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 10 20 30 40 50 60Packet Loss Length (sec)

Probability


Tokyo-Shin Osaka RTT and Packet Loss

0

5000

10000

15000

20000

25000

30000

0 2000 4000 6000 8000 10000

Elapsed time(s)

RT

T(m

s)in-tunnel RTT packet loss

packet loss

Tokyo (0)

Shinagawa (365~443)

Shiyokohama (1084~1168)

Nagoya (6135~6250)

Kyoto (8416~8525)

Shiosaka (9322)


Model of wireless communication environment

Echo CE

Probing CE

Wired Network

CDMA2000 1xEV-DO

Buffer1

Buffer4

Buffer5

Buffer22

Buffer21

Buffer3

Buffer2Echo CE

Probing CE

Wired Network

CDMA2000 1xEV-DO

Buffer1

Buffer4

Buffer5

Buffer22

Buffer21

Buffer3 Buffer1

Buffer2

Echo CE

Probing CE

Wired Network

CDMA2000 1xEV-DO


How to eliminate the buffer function from the network

The original TCP specification had TCP update a smoothed RTT estimator (called R) using the low-pass filter:

R ←− αR + (1 − α)M, α=0.9.

RFC793 recommended the retransmission timeout value (RTO) be set to as follow:

RTO = Rβ, β=2.

From the point of transport layer protocols, a large delay is logically considered to be a link down

We consider that the packet whose RTT is larger than the 2 times of median of RTT is influenced by the buffer1 for the link-down period.


Model of CDMA2000 1xEV-DO communication environment

Buffer1

Echo CE

Probing CE

Wired Network

CDMA2000 1xEV-DO

ARQ BufferBuffer2

Buffer1

Echo CE

Probing CE

Wired Network

CDMA2000 1xEV-DO

ARQ Buffer

Echo CE

Probing CE

Wired Network

CDMA2000 1xEV-DO

Buffer

ARQ Buffer

ARQ Buffer

Buffer

ARQ Buffer


Tokyo-Shin Osaka RTT and Packet Loss(No link-down backup function)

in-tunnel RTT packet loss

0

200

400

600

800

1000

0 2000 4000 6000 8000 10000

Elapsed time(s)

RTT

(ms)

packet loss

Tokyo

(365~443)

Shiyokohama (1084~1168)

Nagoya (6135~6250)

Kyoto (8416~8525)

Shiosaka (9322) (0)

Shinagawa


Cumulative distribution of packet loss length for Tokyo-Shin Osaka

(No link-down backup function)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 10 20 30 40 50 60 70

Probability

Packet Loss Length (sec)


Conclusions and future works

ConclusionsThe route characteristics are introduced in QoS metrics for communication qualities in bullet trains.Traditional two-way ICMP-echo measurement in a specified interval can be used to measure them.A two buffer model-based QoS estimation method is proposed to eliminate the influence by a large buffer in NIC of PC.The raw packet transmission characteristics of CDMA2000 1xEV-DO using the estimation method are gotten.

Future worksThe throughput will be statistically estimated by packet-pair probing. (Preliminary results were got from packets in delay spikes.) Using these parameters, make a model of high speed mobile communication environment on the simulator.

Geneva, 12-13 May 2008

Thank you for your attention!


3GPP2 Mobile IP

HLR

PDSNa

BSCa PCFa

BS

BSC

BS

PCF

RADIUSa

BSC PCFBS

BSCBS

PCF

PDSNb

RADIUSc

HAc

CNMS Access Provider

Network

Home Access Provider Network

Home IP Network (Private)

IPNetwork

HAb

RADIUSb

MS: Mobile StationRAN: Radio Access NetworkBS: Base StationBTS: Base Transceiver StationBSC: Base Station ControllerPCF: Packet Control FunctionPDSN: Packet Data Serving NodeRADIUS: Remote Authentication Dial-In User ServiceFA: Foreign AgentHA: Home Agent

MSC:Mobile Switching CenterHLR:Home Location Register

cdma2000 RAN

BTS MSCA1SS7 Network

Modified from 3GPP2, “cdma2000 Wireless IP Network Standard: Introduction,” X.S0011-001-D, http://www.3gpp2. org/Public_html/specs/X.S0011-001-D_v1.0_060301.pdf, Feb.2006.


(a) CDMA2000 1xEV-DO Protocol Reference Model for MIP4 Control and IKE

cdma2000Air Interface

PPP

IP


PL

A-9

PL

A-9

PL

A-11

MS BTS,BSC PCF

PL

A-11

PL

L2

PDSN/FA

PPP

IP/IPsec

UDP

MIP4

UDP

MIP4 IKE

PL

L2

HA

IP/IPsec

UDP

MIP4IKE



(b) CDMA2000 1xEV-DO Protocol Reference Model for MIP4 User Data

PDSN/FA HA

PL

A-10

PPP

PL

L2

MS BTS,BSC PCF


PPP


PL

A-8

PL

A-8

PL

A-10

IP IP/IPsec

IP

PL

L2

IP IP/IPsec

CN

PL

L2

IP

PL

L2



A-8/A-9 and A-10/A-11

(c) BSC-PCF Interface (d) PCF-PDSN Interface

PL

L2

IP

GRE

PPP

PL

L2

IP

GRE

PPP

PL

L2

IP

UDP

Signalings

PL

L2

IP

UDP

Signalings

(c1) A-8 (c2) A-9 (d1) A-10 (d2) A-11

Documents

ITU-T Kaleidoscope Conference Innovations in NGN TWO BUFFER