Vous êtes sur la page 1sur 18

IP networks - throughput

Magnus Abrahamsson 2010-02-17

Mid Sweden University


The Department of Information Technology and Media (ITM)
Author: Magnus Abrahamsson
E-mail address: mrabris@gmail.com
Scope: 2223 words inclusive of appendices
Date: 2010-02-16
IP networks - throughput
Magnus Abrahamsson 2010-02-17

Report
Computer A, Computer networks, 7.5 points,
DT024G (distance)

IP networks - throughput
Test tools

Magnus Abrahamsson
IP networks - throughput
Magnus Abrahamsson 2010-02-17

Foreword
This is a short technical report written as an assignment in the course
Datornätverk. The report is actually an extension to the one of the topics
we studied in this course. I have chosen write a report that you could
say is a mix of the subject “Test and compare two types of software in the
area of data communication” and “Compare two data communication stan-
dards”. I will look closer on test tools to measure IP network perform-
ance with focus on throughput, and describe their advantages and
disadvantages.

It has been quit a challenge for me, but a good one. I hope you enjoy
reading it as much as I liked writing and working with it.

Kind regards

Magnus Abrahamsson
IP networks - throughput
Magnus Abrahamsson 2010-02-17

Table of Contents
Foreword ..........................................................................................................iii

1 Introduction............................................................................................5
1.1 Scope .............................................................................................5
1.2 Outline ..........................................................................................5
1.3 Contributions ...............................................................................5

2 Network throughput introduction.....................................................6


2.1 Throughput ..................................................................................6
2.2 Goodput (data transfer rate)......................................................7

3 Network tools.........................................................................................8
3.1 iperf ...............................................................................................8
3.2 jperf................................................................................................8
3.3 ntop................................................................................................9

4 Test tool measurements......................................................................11


4.1 Testbed........................................................................................11
4.2 iperf .............................................................................................12
4.2.1 TCP throughput test.......................................................12
4.2.2 Testing UDP ....................................................................12
4.2.3 Simulate application data traffic ..................................13
4.2.4 Bi-directional test............................................................14
4.3 ntop..............................................................................................15
4.3.1 Network traffic historical view.....................................15
4.3.2 Network Throughput ....................................................16

5 Conclusions / Discussion...................................................................17

6 References.............................................................................................18
IP networks - throughput Introduction
Magnus Abrahamsson 2010-02-17

1 Introduction
How to measure throughput and if it’s really delivered is one of the
topics both end-users and network operators are discussing right now.
As networks become larger, more complex, and more heterogeneous
network management turns into an increasingly complex task. It re-
quires different types of network monitoring and diagnostic tools then
ping and traceroute, how are suitable just for tackling simple connectivity
problems. Both automated tools and special performance tools will be
necessary in the future to support the human effort.

This document is intended to provide valuable information and


examples for network managers or operators on the use of two different
network tools; iperf and ntop.

1.1 Scope
There are several different network monitoring and diagnostic tools and
ways to measure the network performance in an IP network. This study
has its focus on how to measure of bandwidth/throughput with iperf and
ntop.

1.2 Outline
Chapter 2 gives you a description of different IP network performance
terms.

Chapter 3 gives you information about network test & measurement


tools

Chapter 4 gives you more hands-on examples of about the tools.

Chapter 5 gives you conclusions and more comments on the subject.

1.3 Contributions
Special big thanks to all colleagues at work which have been very sup-
portive during the whole process.
IP networks - throughput Network throughput introduction
Magnus Abrahamsson 2010-02-17

2 Network throughput introduction


Before we look into what kind of data the different test tools can provide
us with, it’s impotent to be able to distinguish between some network
terms associated with throughput. Therefore I will try to walk you
through some of them in this chapter.

2.1 Throughput
How fast we actually can send data through a network. It’s usually
measured bit per second (bps).

You have to keep in mind that the value of “maximum throughput” could
have four different meanings depending of its context. They are:

Type of throughput Description

Maximum theoretical throughputBased on ideal circumstances.


Primarily used as a rough
calculated value to determining
bounds in a design phase.
Maximum Achievable throughput Maximum data rate of successful
data transfer through a communica-
tion path.

Peak measured throughput (peek) Measured on real transaction for a


short period of time. Useful for
system that relies on burst data
transactions. For system with high
duty time it less likely of interest.

Maximum sustained throughput The average or integrated over long


(avg) time throughput. God for high duty
networks to measure the perform-
ance. Could be misleading if pack-
age shaping are used.

Current measured throughput Real transaction in progress meas-


IP networks - throughput Network throughput introduction
Magnus Abrahamsson 2010-02-17

(current) ured throughput.

Table 1: Type throughputs

Keep an eye on the lowest value link in the series, it’s referred to as the
bottleneck.

2.2 Goodput (data transfer rate)


Goodput or data transfer rate refers to the achieved average net bit rate
that is delivered to the application layer, exclusive of all protocol
overhead, data packets retransmissions, etc.

For example, in the case of file transfer, the goodput corresponds to the achieved
file transfer rate. The file transfer rate in bit/s can be calculated as the file size
(in bytes), divided by the file transfer time (in seconds), and multiplied by eight.

If no data compression is provided by the network equipment or


protocols, we have the following relation:

Goodput ≤ Throughput ≤ Maximum throughput ≤ Net bit rate


IP networks - throughput Network tools
Magnus Abrahamsson 2010-02-17

3 Network tools
We will in this chapter look deeper into two different tools, and see
what kind of network performance they actually measures.

3.1 iperf
iperf is a nifty little program for measuring TCP and UDP performance
between end points. It has both client and server pieces, so it requires
installation at both ends of the connection you're measuring. Via this
tool you then could measure throughput on your various network
segments, and collect jitter and datagram loss statistics. [4]

Primarily Measured Throughput:


Maximum Achievable throughput

iperf Advantages
+ Easy to install
+ Possible to simulate application traffic
+ Could be running directional or bi-directional
+ Measures jitter on simulated traffic (UDP)
+ Measures packet lost.
+ Free (Open-source)

iperf Disadvantages
- lack of documentation

3.2 jperf
It’s possible to use a Java based GUI for iperf called jperf. I not going into
any further detail regarding this tool, only show you a screenshot from
it.
IP networks - throughput Network tools
Magnus Abrahamsson 2010-02-17

Picture – jperf GUI

3.3 ntop
ntop is a wonderful hybrid packet analyzer that generates nice clickable
HTML reports that show you what's happening on your network. It
slices and dices network traffic all kinds of ways: by protocol, host, local
or remote network, network load, network flow, what Web sites your
users are visiting, how much traffic is coming from or going to remote
sites, throughput and loads more. It supports virtually all network
protocols over both IP networks and Fibre Channel. [5]

Primarily Measured Throughputs:


Maximum sustained throughput (avg)
Peak measured throughput (peek)
Current measured throughput (current)

ntop Advantages
+ Easy to install
+ Simple application traffic
+ Free(Open-source)
+ Traffic measurement
+ Traffic monitoring
+ Network optimization and planning
IP networks - throughput Network tools
Magnus Abrahamsson 2010-02-17

+ Detection of network security violations

ntop Disadvantages
- (sampling period)
IP networks - throughput Test tool measurements
Magnus Abrahamsson 2010-02-17

4 Test tool measurements


In this chapter we will give actual examples on how the two test tools,
iperf and ntop, worked in a live environment.

4.1 Testbed

Server1
Redhat HP server, 2 Dual-Core AMD Opteron(tm) Processor 2218
processors (4 cpu cores) (version 2.00.00-rhel4), 4Gb memory
1000 Mbps full duplex, receive & transmit flow control ON

Broadcom NetXtreme II BCM5708 1000Base-T (B2) PCI-X 64-bit 133MHz

Client1
Ubuntu 9.10 desktop, Pentium 4 3Ghz, 1Gb memory
100 Mbps, full duplex. Flow control is off for TX and off for RX.

Tigon3 [partno(BCM95751) rev 4001 PHY(5750)] (PCI Express)


10/100/1000Base-T Ethernet

ntop Iperf iperf


server client server

Client1 Server1

Subnet A Subnet B
Sw R Sw
IP networks - throughput Test tool measurements
Magnus Abrahamsson 2010-02-17

4.2 iperf
First start up iperf in server mode with this command:
admin@server1# iperf -s

Run the client with this command:


guest@client1:~$ iperf -c server1

4.2.1 TCP throughput test


By default iperf uses TCP port 5001, so make sure it's not blocked. This
is the result of a run without tcpdump running:
-----------------------------------------------------
---
Server listening on TCP port 5001
TCP window size: 85.3 KByte (default)
-----------------------------------------------------
---

[ 3] 0.0-10.0 sec 113 MBytes 94.6 Mbits/sec

Pretty nice- that's as good as you can get on Fast Ethernet. This is what
happens when tcpdump is running on server side:

[ 4] 0.0-10.0 sec 110 MBytes 92.0 Mbits/sec

4.2.2 Testing UDP


Stop the server with Ctrl+C, then run these commands to start iperf
server in UDP mode.

admin@dserver1:~$ iperf -su

from the client run the following:

guest@client1:~$ iperf -c deneb –u


[ ID] Interval Transfer Bandwidth
[ 3] 0.0-10.0 sec 1.25 MBytes 1.05 Mbits/sec
[ 3] Sent 893 datagrams

[ 3] Server Report:
[ 3] 0.0-10.0 sec 5638046520444870 bits 0.00
(null)s/sec 0.007 ms 0/ 893 (0%)

[ 3] 0.0-10.0 sec 1.25 MBytes 1.05 Mbits/sec


0.003 ms 0/ 893 (0%)
IP networks - throughput Test tool measurements
Magnus Abrahamsson 2010-02-17

That's quite a difference. Why is UDP so slow? Because iperf's default is


1.05 Mbits/second, so it's not a network problem. We can try some
different values to see what happens. Let's tell it to use all available
bandwidth:

admin@server1# iperf -su

guest@client1# iperf -c server1 -u -b 100m


Client connecting to server1, UDP port 5001
Sending 1470 byte datagrams
UDP buffer size: 110 KByte (default)
----
[ ID] Interval Transfer Bandwidth
[ 3] 0.0-10.0 sec 114 MBytes 95.6 Mbits/sec
[ 3] Sent 81314 datagrams
[ 3] Server Report:
[ 3] 0.0-10.0 sec 513169564593235200 bits 0.00
(null)s/sec 0.009 ms 33/81313 (0.041%)
[ 3] 0.0-10.0 sec 1 datagrams received out-of-
order

That's very good speed, and 0.041% datagram loss is insignificant. That's
a good clean connection. VoIP call can tolerate as much as 10% UDP
datagram loss.

4.2.3 Simulate application data traffic


Applications determine how many TCP or UDP packets are sent, and
what size. To get a more real-world idea of performance, you can set the
size of the UDP datagram to the same size that your applications use.

This example whales on your line by sending 200-byte datagrams at 100


Mbits/second:

admin@server1:~$ iperf -su -i 1


guest@client1:~$ iperf -c server1 -u -l 200 -b 100m
[ID] Interval Transfer Bandwidth
[3] 0.0-10.0 sec 89.0 MBytes 74.7 Mbits/sec 0.054
ms 4728/466661 (1%)

The -i option generates a progress display every second.


IP networks - throughput Test tool measurements
Magnus Abrahamsson 2010-02-17

4.2.4 Bi-directional test


By default, iperf flings TCP packets over your wires as fast as possible.
A bi-directional test, which is the -d option, runs both ways:

guest@client1:~$ iperf -c server1 -d


[ ID] Interval Transfer Bandwidth
client1 -> server1
[ 5] 0.0-10.0 sec 48.8 MBytes 40.9 Mbits/sec
Server1 -> client1
[ 4] 0.0-10.0 sec 110 MBytes 91.8 Mbits/sec

As you can see the Client1 machine has big troubles with sending
packages to the server during bidirectional testing. Why kind of
bottleneck is this
?
- This is because Clinet1 hasn’t flow control turn on?
- Or is it because of the 1000Base-T and 100Base-T differ?
- Or could it be the swiches on subnet A?

Let do a bi-directional UDP test and see what happens:

guest@client1:~$ iperf -c server1 -u -b 100m –d


[ 4] 131.116.248.17 > 131.115.4.164 port 5001
[ 3] 131.115.4.164 > 131.116.248.17 port 5001
[ ID] Interval Transfer Bandwidth
[ 4] 0.0-10.0 sec 114 MBytes 95.6 Mbits/sec
[ 3] 0.0-10.1 sec 114 MBytes 95.1 Mbits/sec
0.029 ms 3891/85459 (4.6%)

We can see that we now have almost no jitter (0.029ms) but a lot of
package lost (4.6%) in the other direction, server -> client
communication.
IP networks - throughput Test tool measurements
Magnus Abrahamsson 2010-02-17

4.3 ntop
We will only look at the ntop in web-mode.

You can start the ntop server with:


#/etc/init.d/ntop start
You can access the ntop server webpage on default port 3000.
http://hostname:3000

4.3.1 Network traffic historical view


The network traffic historical view gives you a lot of information about
the network traffic. It’s easy to see what kind of traffic there is in the
network and its average/peek values during the period.

Table3 - Historical View per service


IP networks - throughput Test tool measurements
Magnus Abrahamsson 2010-02-17

4.3.2 Network Throughput


Those are graphs that show the evolution of the total throughput
observed in the network. They are presented in different time scales,
showing the throughput in the last 10 minutes up to last month. This
sort of statistics is valuable to determine peek and low usage periods. In
this way the administrator will be able to better schedule traffic
intensive or network disruptive activities (physical network
maintenance, switch configuration, data traffic with low priority, etc.). It
might also be interesting to detect unexpected throughput peeks, which
could indicate excessive use of the network resources by a user or group
of users. [5]

It’s important to understand that ntop use minimum sampling period of


a minute. This means that the throughput peak value actually is the
peak average of a minute during the measurement period. In the most
cases this is accurate enough for network monitoring.

Table4 – Network Throughput : Data sent + revceived

Table5 –Last hour Network Throughput (1 min sampling period)


IP networks - throughput Conclusions / Discussion
Magnus Abrahamsson 2010-02-17

5 Conclusions / Discussion
Measuring the throughput, delay, jitter and packet loss only gives you a
hint of where you could have performance issues or the bottleneck in
the network. Finding network problems does not necessarily lead to
instant solutions. Correct interpretation is still needed. Sometimes it can
be frustrating to have a lot of information without finding the answer to
your problems. However, network monitoring can truly be an asset to
your corporation as a whole.
IP networks - throughput References
Magnus Abrahamsson 2010-02-17

6 References

[1] By Behrouz A. Forouzan; Data Communications and Networking, 4th


edition; McGraw-Hill, 2007. ISBN 007-125442-0.

[2] By Craig Hunt; TCP/IP Network Administration. O’Reilly. ISBN 0-


937175-82, s257-300, May 1994

[3] By Gian-Paolo D. Musumeci & Mike Loukides; System Perform-


ance Tuning, 2nd edition ; O’Reilly, Feb 2002, s86-91. ISBN 0-201-
41979-3

[4] By Carla Schroder, Linux Network cookbook, O’Reilly. ISBN 10: 0-


596-10607-6. November 2007

[5] [Deri98] Deri, L. NTOP User’s Guide - Network Usage Monitor for
Unix Systems.Centro Serra, University of Pisa, Italy. Available at
http://www.ntop.org/ntop-overview.pdf. Fetched 2010-02-16

Front page Illustration: IPv4 INTERNET TOPOLOGY MAP, By UC


Regents 2008, AS level Internet graph.

Vous aimerez peut-être aussi