COM 12 – D 156 – E
INTERNATIONAL TELECOMMUNICATION UNION
TELECOMMUNICATION
STANDARDIZATION SECTOR
English only
STUDY PERIOD 2001-2004
Original: English
Questions:
12 & 13/12
Geneva, 22-30 September 2003
STUDY GROUP 12 – DELAYED CONTRIBUTION 156
Source:
Nortel Networks (Canada)
Title:
Estimating end to end performance in IP networks
1.
Introduction
Methods for estimating user acceptance of voice performance in end to end telecommunications
networks are well known, for example by the use of the Rec. G.107 E-model and the planning
guidelines given in Rec. G.108. However, with the increasing amount of non-voice traffic on the
network, there is a need to provide a similar analysis capability to determine end-user acceptance.
One of the big differences to take into account as compared to voice networks are the use of
different protocols and the effect of impairment accumulation. The aim of this contribution is to
show a simple method for estimating end to end performance in IP networks, and illustrate with
examples.
2.
Reference connection
For the purposes of this contribution, the generalised reference connection shown in Figure 1 is
used, where an end user accesses an application or server via an access network and a core network.
The overall end to end user performance is then a function of the performance of the individual
elements.
End to end performance
Access network
performance
Access
Network
Core Internet
performance
End user/server
performance
Managed
IP network
End-user/
application
server
DNS
server
Figure 1: Reference connection
Contact:
Paul Coverdale
Nortel Networks
Canada
Tel:
+1 613 763 4277
Fax:
+1 613 765 7723
Email: paulcov@nortelnetworks.com
Attention: This is not a publication made available to the public, but an internal ITU-T Document intended only for use by the
Member States of the ITU, by ITU-T Sector Members and Associates, and their respective staff and collaborators in their ITU related
work. It shall not be made available to, and used by, any other persons or entities without the prior written consent of the ITU-T.
-2COM 12 – D 156 – E
3.
Impairment accumulation rules
To provide an end to end performance analysis, a simple set of impairment allocation rules is
proposed. It is recognised that these rules are not rigorous, but it is felt that they provide a
reasonable basis for estimating overall performance. The parameter definitions follow those in
Recs. Y.1540 and Y.1541
1.
Mean delays (IPTD) for individual elements are directly additive, e.g. 100 ms + 50 ms =
150 ms total
2.
Packet loss ratios (IPLR) for individual elements are directly additive, e.g. 0.1% + 0.01% =
0.11% total
3.
Delay variations (IPDV) for individual elements are additive on an RMS basis (i.e.
variances are additive), e.g. 10 ms + 20 ms = (102 + 202) = 22.36 ms total
4.
Following the Y.1541 definition of IPDV, the effective contribution of the total IPDV to
the overall delay budget is ½ IPDV
4.
Application examples
The following examples are used to illustrate how the impairment accumulation rules above can be
used to estimate end to end performance in an IP network.
4.1
Web-browsing
The protocol exchange of HTTP, as used for web-browsing, is shown in Figure 2, and involves
several back and forth exchanges.
User
DNS response time
DNS lookup
Total time
apparent
to user
Server
Open TCP
Web-server
response time
HTTP GET
Transmission
time
Page
download
Time
Time
Figure 2: HTTP protocol exchange
The total time apparent to the user for downloading a web-page is given by:
Tuser = Tdnslookup + Ttcpconnect + Thttpget + Tserver response + Tdownload time
This can be written as:
Tuser = (2Tacc + Tdns) + 2(Tacc + Tint) + 2(Tacc + Tint) + Tserver + Ttxfr
= 6 x Tacc + 4 x Tint + Tdns + Tserver + Ttxfr
-3COM 12 – D 156 – E
where:
Tacc = Access network delay (including jitter)
Tdns = DNS server response time
Tint = Core internet delay (including jitter)
Tserver = Web server response time
Ttxfr = Data transfer time, given by file size/bit-rate
Separating the access and core internet delay into IP packet delay and delay variation, this can be
written as:
Tuser = 6 x IPTDa + 4 x IPTDc + Tdns + Tserver + ½(6 x IPDVa2 + 4 x IPDVc2) + Ttxfr
where:
IPTDa = IP packet transfer delay for the access network
IPTDc = IP packet transfer delay for the core internet
IPDVa = IP packet delay variation for the access network
IPDVc = IP packet delay variation for the core internet
4.1.1
Numerical examples
For illustration purposes, the following combination of parameter values are used:
Core network:
IPTDc = 100 ms, IPDVc = 50 ms
IPTDc = 400 ms, IPDVc = 100 ms
Access network:
IPTDa = 50 ms, IPDVa = 20 ms
IPTDa = 100 ms, IPDVa = 50 ms
The bit-rate achievable on the access network is assumed to be the limiting bit-rate for the complete
end to end connection.
Web server response time = 180 ms
DNS server response time = 20 ms
Using these values, the time apparent to the user (Tuser) for downloading a web-page, as a function
of file size and bit- rate is given in the tables below.
File size = 10 KB, bit rate = 56 kb/s
IPTDc = 100 ms
IPDVc = 50 ms
IPTDc = 400 ms
IPDVc = 100 ms
IPTDa = 50 ms
IPDVa = 20 ms
2.4s
3.7s
IPTDa = 100 ms
IPDVa = 50 ms
2.7s
4.0s
-4COM 12 – D 156 – E
File size = 100 KB, bit rate = 56 kb/s
IPTDc = 100 ms
IPDVc = 50 ms
IPTDc = 400 ms
IPDVc = 100 ms
IPTDa = 50 ms
IPDVa = 20 ms
15.3s
16.5s
IPTDa = 100 ms
IPDVa = 50 ms
15.6s
16.8s
File size = 100 KB, bit rate = 1 Mb/s
IPTDc = 100 ms
IPDVc = 50 ms
IPTDc = 400 ms
IPDVc = 100 ms
IPTDa = 50 ms
IPDVa = 20 ms
1.8s
3.0s
IPTDa = 100 ms
IPDVa = 50 ms
2.1s
3.3s
By way of comparison, Rec. G.1010 suggests a preferred target of <2s for downloading a web-page
(~ 10KB), with an acceptable target of < 4s. In this case, the file size and bit-rate have a dominant
effect, and the advantage of high speed access in meeting user expectations can clearly be seen.
5.
Example for email
The protocol exchange of POP3, as used for downloading email, is shown in Figure 3, and again
involves a number of back and forth exchanges.
User
Server
DNS response time
DNS lookup
Open TCP
Server
responds
User
name
Password
Total time
apparent
to user
STAT
UIDL
LIST
Retrieve
mail
Message
download
Transmission
time
Time
Time
Figure 3: POP3 protocol exchange
-5COM 12 – D 156 – E
In this case, the user is presumed to have already logged into the system, so that no additional time
for DNS lookup is required. In this case, the time apparent to the user is given by:
Tuser = 16(Tacc + Tint) + Ttxfr
where:
Tacc = Access delay (including jitter)
Tint = Core internet delay (including jitter)
Ttxfr = Data transfer time, given by file size/bit-rate
Separating the access and core internet delay into IP packet delay and delay variation, this can be
written as:
Tuser = 16 x IPTDa + 16 x IPTDc + ½(16 x IPDVa2 + 16 x IPDVc2) + Ttxfr
where:
IPTDa = IP packet transfer delay for the access network
IPTDc = IP packet transfer delay for the core internet
IPDVa = IP packet delay variation for the access network
IPDVc = IP packet delay variation for the core internet
5.1
Numerical examples
For illustration purposes, the following combination of parameter values are used:
Core network:
IPTDc = 100 ms, IPDVc = 50 ms
IPTDc = 400 ms, IPDVc = 100 ms
Access network:
IPTDa = 50 ms, IPDVa = 20 ms
IPTDa = 100 ms, IPDVa = 50 ms
The bit-rate achievable on the access network is assumed to be the limiting bit-rate for the complete
end to end connection.
Using these values, the time apparent to the user (Tuser) for downloading an email, as a function of
file size and bit- rate, is given in the tables below.
File size = 10 KB, bit rate = 56 kb/s
IPTDc = 100 ms
IPDVc = 50 ms
IPTDc = 400 ms
IPDVc = 100 ms
IPTDa = 50 ms
IPDVa = 20 ms
3.9s
8.7s
IPTDa = 100 ms
IPDVa = 50 ms
4.8s
9.7s
-6COM 12 – D 156 – E
File size = 10 KB, bit rate = 1 Mb/s
IPTDc = 100 ms
IPDVc = 50 ms
IPTDc = 400 ms
IPDVc = 100 ms
IPTDa = 50 ms
IPDVa = 20 ms
2.6s
7.5s
IPTDa = 100 ms
IPDVa = 50 ms
3.4s
8.3s
By way of comparison, Rec. G.1010 suggests a preferred target of <2s for downloading an email
(~ 0KB), with an acceptable target of < 4s. In this case, due to the large number of back and forth
protocol exchanges, the delays in the access and core network have a more dominant effect than
bit-ate.
6.
Summary and conclusions
The analysis shown here provides a simple method of estimating end to end performance in IP
networks from a user perspective. The method can be used to analyse the performance of specific
connection scenarios and as a tool to assist in developing realistic performance objectives for
individual elements. It is recommended to use this Contribution to form the basis of a new SG12
Recommendation.
___________________