Professional Documents
Culture Documents
Quality of Service in Network Administrations: Abstract
Quality of Service in Network Administrations: Abstract
Abstract:
Today’s computer networks have gone from typically being a small local area network, to wide
area networks, where users and servers are interconnected with each other from all over the world. This
development has gradually expanded as bandwidth has become higher and cheaper.But when dealing
with the network traffic, bandwidth is only one of the important properties. Delay, jitter and reliability are
also important properties for the quality of network connection. This is because different applications has
different needs, and therefore require different properties from the network. System administrators are in
an increasing degree involved with the troubleshooting of solving network problems concerning the
quality of service for the different applications.This research paper analyzed techniques for measuring,
analyzing, presenting and interpreting the different properties for the administration of remote computer
network.
written in terms of the reliability function, R(t), measuring these and more properties will be
as, shown.
Case One: Network Traffic
Motivation
By monitoring the network traffic for a node,
where, t1 and t2 are the beginning and ending information about the state of that node can be
of a specified interval of time, and R(t) is the determined. This may provide the network
reliability function, i.e. probability of no failure administrator, with enough information to
before time t. optimize the system performance, by removing
The failure rate data can be obtained in bottlenecks. The system can represent the node,
several ways. The most common methods are a subnet, or the whole network.
[30]: There are especially two locations that are of
• Historical data about the device or system interest, when performing passive network
under consideration. measurements:
• Government and commercial failure rate • The state of a service host, that provides a
data. network service. Examples of network
• Testing. services are the DNS, DHCP, HTTP, and
When monitoring a network connection FTP services.
or a node, packet loss is a measurement for • The state of a network node, performing a
measuring the fraction of packets sent from a routing or forwarding functions. Examples of
measurement node to a destination node for such nodes are firewalls, virtual private
which the measurement node does not receive an networks (VPN), and routers.
acknowledgment from the destination node [30]. Objective:
Data Collection By using a passive network measurement
The active measurement tools used for tool, two nodes are to be monitored for one day.
measuring the delay and jitter, are also suitable The data gathered from these nodes are to be
for measuring the failure rate. analyzed, and the state of the nodes is of interest.
Analysis Resources:
The interesting information gathered from In this experiment, the resources located in table
the raw data conserving the reliability, is how 1.4 have been utilized.
many error there where during the measurement Table 1.4: The resources utilized in Case One.
period. This is known as the error rate. It shows Description Node One Node Two
the number of errors divided on the time
Processor Intel AMD Athlon(tm)
interval. Processor Mhz 549,947Mhz 1852,314Mhz
To retrieve reliable failure rate data, the Memory 640 MB 1024 MB
testing should be performed over a relative large Network MAC Fast Ethernet Ethernet FD
period of time. This removes uncertainty in the Network Link 100 Mb/s 10 Mb/s
result. Internet Service UNINETT Bredbandsbolaget
IP Address 128.39.73.19 83.227.111.133
Table 1.5: The resources utilized in Case Two Debian, where the version of Debian is
Description Node One Node Two Node "Woody". The host is also running the
Processo Intel P III Intel P Intel P III following services: MYSQL, NTP, SMTP,
Processor Mhz 549,947M 167,047M 447,699M HTTP, FTP, and SSH.
Memory 640 MB 96 MB 923 MB The host is connected to a local area
Network MAC Fast Fast Fast network, which shares a 155Mb/s internet
Network Link 100Mb/s 100Mb/s 100Mb/s connection with the rest of the school. But as
Internet Service UNINETT UNINETT UNINETT the local area network, and the host network
IP Address 128.39.73. 158.38.88. 128.39.74.
interface card only supports Fast Ethernet, the
19 147 16
The host is connected to a local area maximum network speed is about 100Mb/s.
network, which shares a 155Mb/s internet The path from Node One to
connection with the rest of the school. But as node two is shown in table 1.7.
the local area network, and the host network Table 1.7: Path from Node One to Node Three.
interface card only supports Fast Ethernet, the Path ID IP Location
maximum network speed is about 100Mb/s. 01 02 128.39.73.1 Oslo, Norway
Description of Node Two 128.39.74.16 Oslo, Norway
Node Two is a host located in the student
housings at Molde University College, in
Tools
Norway. The operating system is GNU Linux
There are multiple tools that perform about
Red Hat, where the version of Red Hat is
the same function when performing active
"9.0".
measurements. Known network throughput
The host is connected to a local area benchmarking tools are: netperf, iperf, ttcp, and
network, which shares a 155Mb/s internet ftp.
connection with the rest of the school. But as
the local area network, and the host network The tool chosen to test the throughput is
interface card only supports Fast Ethernet, the netperf. To execute the experiment, a server
maximum network speed is about 100Mb/s. node and a client node has to be installed on
The path from Node One to Node Two is each of the nodes.
shown in table 1.6. For the experiment, the server program was
Table 1.6: Path from Node One to Node Two. installed on Node Two and three, and the client
Path ID IP Location software was installed on Node One. This setup
was chosen, so that the process of benchmarking
01 128.39.73.1 Oslo, Norway
02 158.36.84.21 Oslo, Norway could be controlled from Node One. This
03 128.39.0.73 Oslo, Norway minimizes the probability for interference from
04 128.39.46.249 Oslo, Norway each measurement.
05 128.39.46.2 Trondheim, Norway Predictions
06 128.39.46.102 Trondheim, Norway The predictions for the result are as follows:
07 128.39.47.102 Ålesund, Norway • As all nodes are attached to an over
08 128.39.47.130 Molde, Norway dimensioned network, the connection itself
09 158.38.0.66 Molde, Norway should not be a problem. And it should be
10 158.38.88.147 Molde, Norway possible to achieve full link utilization.
• Node Two could have a problem to achieve
Description of Node Three 100 Mb/s as the processing power is a bit
Node Three is a host located in the student low.
network at Oslo University College, in Oslo • All the nodes are connected to a school
Norway. The operating system is GNU Linux network. This will probably mean that the
ISSN: 2395-1303 http://www.ijetjournal.org Page 135
International Journal of Engineering and Techniques - Volume 3 Issue 3, May- June 2017
link has the highest load during the day. This In addition, the link and processing power of
is why there is a higher chance to achieve full three remote nodes has been utilized. As the
link utilization during the night or weekends. active measurements does not require any
installation or configuration of the destination
Case Three: Delay, Jitter and Packet Loss nodes, the hardware configuration of Node
Motivation Three and four are not known. The hardware
• By using an active measurement tool that configuration of Node Two can be found in table
measures the delay between two nodes, the 1.5 (Node Two).
jitter and packet loss can be determined by Description of Node One
using mathematical methods. Node one is a host located in a testlab at
• The delay can be measured as the time it Oslo University College, in Oslo Norway. The
takes for one packet to be sent from a host, operating system is GNU Linux Debian, where
until it is received at the destination. But as the version of Debian is "Sarge". The host is also
this requires that the clocks are perfectly running the following services: PostgreSQL,
synchronized, an alternative method is SMTP, HTTP, HTTPS, FTP, and SSH.
mostly used. This is to measure the delay in
form of the round trip time. The host is connected to a local area
• The round trip time is measured as the time it network, which shares a 155Mb/s internet
takes for one packet to be sent from a node to connection with the rest of the school. But as
a destination node, until another packet is the local area network, and the host network
received from the destination node. interface card only supports fast ethernet, the
Objective maximum network speed is about 100Mb/s.
• The previous cases showed methods for Description of Node Two
measuring the throughput for the link. In this Node Two is a host located in the student
last case study, the delay of a network link housings at Molde University College, in Molde
is to be measured, and based on those Norway. The operating system is GNU Linux
measurements; the jitter and packet loss is to Red Hat, where the version of Red Hat is "9.0".
be determined. The host is connected to a local area
• The round trip time, from one node to three network, which shares a 155Mb/s internet
other nodes are to be measured for one week. connection with the rest of the school. But as
This should provide enough information to the local area network, and the host network
make reasonable decisions about the link interface card only supports Fast Ethernet, the
state. maximum network speed is about 100Mb/s.
Resources
Description of Node Three
In this experiment, the resources located in
Node Three is a node that is a part of a
table 1.8 have been utilized.
cluster that serves the "www.vg.no" domain.
Table 1.8: The resources utilized in Case
This domain belongs to a Norwegian news paper
Three.
called "Verdens Gang". The path from Node
Description Node One
One to Node Three can be view in table 1.9.
Processor Model Intel Pentium III
Processor Mhz 549,947Mhz
Memory 640 MB
Network MAC Fast Ethernet FD
Network Link 100 Mb/s
Internet Service Provider UNINETT
IP Address 128.39.73.19
Table 1.9: Path from Node One to Node Three. round trip times from links all around the world.
Path ID IP Location The modified Perl scripts together with other
01 02 128.39.73.1 Oslo, Norway scripts are available in the appendix.
03 04 158.36.84.21 Oslo, Norway Predictions
05 06 128.39.0.73 Oslo, Norway The predictions for the result are as follows:
07 08 193.156.120.3 Oslo, Norway • Node Three has the least amount of hops,
193.75.3.6 Oslo, Norway and is located in Oslo, Norway, this node
193.75.3.1 Oslo, Norway will probably have the lowest delay, the
193.69.165.11 Oslo, Norway lowest jitter, and the lowest packet loss.
193.69.165.11 Oslo, Norway • Node Two has the second least amount of
hops, and it is located in Norway, so this
Description of Node Four node will probably have a low delay, a
Node Four is a node that is a part of a cluster that predictable jitter, and a low packet loss.
serves the "www.kernel.org" domain. This • Node four is located in the United States of
domain belongs to the official GNU Linux America. This will probably result in a high
kernel. The path from Node One to node four delay, with at times unpredictable jitter. The
can be view in table 1.10. packet loss should however be relative low,
Table 1.10: Path from Node One to Node Four. with today’s network link properties.
Path ID IP Location
01 128.39.73.1 Oslo, Norway CONCLUSION
02 158.36.84.21 Oslo, Norway The objective is to assist network and
03 128.39.0.73 Oslo, Norway system-administrators in administration of
04 128.39.46.249 Oslo, Norway remote computer networks.
05 193.10.68.101 Oslo, Norway This was primary done by identifying the
06 193.10.68.29 Stockholm, Sweeden properties for securing the remote computer
07 213.242.69.21 Stockholm, Sweeden networks, and the properties that are important for
08 213.242.68.201 Stockholm, Sweeden the quality of the services. The properties provide
09 212.187.128.25 London, England quality of service for the connection between the
10 4.68.128.106 New York, USA remote computer networks.
11 64.159.1.130 San Jose, USA Secondary, some simple methods for
12 4.68.114.158 San Jose, USA analyzing and presenting the measured data was
13 209.245.146.251 San Jose, USA identified. These methods simplify the
14 192.5.4.233 San Jose, USA interpretation part of the administration of
15 204.152.191.5 San Jose, USA remote computer networks.
The three case studies were created to
Tools
demonstrate the functionality for some of the
There are multiple tools for measuring the
tools used to measure the four properties in
round trip time, or one-way delays. The tools
quality of service.
vary in methods for collecting and measuring the
In Case One, the objective was to make use
data. The biggest difference is what sort of
of passive throughput measurement tools, to
packet which is used. The available packet
monitor the traffic on two different nodes for
formats can be ICMP, TCP or UDP.
one day. The tcpstat tool successfully measured
The tool used to measure the data, is a the data on both nodes, and provided enough
modified Perl script that utilized the ping information to create a good understanding of
command which is available for most operating what had happened on the network for the last
systems. The Perl script is a part of the pinger 24 hours. The only mistake in these two
measurement package, which is used to measure experiments was that the filter functionality in
tcpstat should been used to filter input and [3] Annabel Z. Dodd. The Essential Guide to
output traffic. But as the objective was Telecommunications, Second Edition.
demonstrate, the experiments can still be Prentice Hall PTR, 1999.
classified as successful. [4] Sergio Verdú. Wireless bandwidth in the
In Case Two, the objective was to make use making. IEEE, 2000.
of active throughput measurement tools, to [5] Gene Spafford Simson Garfinkel, Alan
benchmark the network connection between two Schwartz. Practical Unix & Internet
different nodes for one week. The netperf tool Security, 3rd Edition. O’Reilly, 2003.
successfully measured the throughput from one [6] Fred Halsall. Data Communications,
node, to two other nodes situated at different Computer Networks and Open Systems,
computer networks. In this case, the results did Fourth Edition. Addison-Wesley, 1996.
not match the predictions, and this had probably [7] William Stallings. Local networks. ACM,
something to do with limited hardware resources 1984.
at the nodes. It would have been interesting to [8] Mahbub Hassan and Raj Jain. High
remove that bottleneck, and have really tested the Performance TCP/IP Networking. Pearson
network throughput between two high Prentice Hall, 2004.
performance nodes, or at least include the CPU of [9] W. Richard Stevens. The Protocols (TCP/IP
the measurement nodes during the measurement. Illustrated, Volume 1). Addison-Wesley,
But again the objective was to demonstrate the 1993.
active throughput tools, and since netperf [10] Matt Bishop. Computer Security, Art and
performed as expected, the experiment can still Science. Addison-Wesley, 2002.
be classified as successful. [11] Cross-Industry Working Team. Internet
service performance: Data analysis and
In Case three, the objective was to make use visualization. Technical report, The Cross-
of active delay measurement tools, to benchmark Industry Working Team (XIWT), 2000.
the network connection between three different [12] Mark Burgess. Analytical Network And
nodes for one week. The round trip time System Administration. Wiley, 2004.
measurement that was done in the experiment [13] B. Kleiner P.A. Tukey J.M. Chambers,
can be used to find both the delay and the jitter of W.S. Cleveland. Graphical Methods for
a connection. These three properties provide Data Analysis. Duxbury Press, 1983.
important information about the quality of service [14] Lionel M. Ni Xipeng Xiao. Internet qos - a
for the connection. And with the available big picture. IEEE, 1999.
information relative important assumptions could [15] Bruce S. Davie Larry L. Peterson.
be done to describe the quality of service for Computer Networks - A System Approach,
these network connections. This helps the Second Edition. Morgan Kaufmann, 2000.
system administrator in designing the network [16] Joseph D. Sloan. Network Troubleshooting
for services that require different quality of Tools. O’Reilly, 2001.
service properties. This experiment can [17] Rick Jones. The netperf website.
definitely be classified as successful. http://www.netperf.org/, May 2005.
[18] The Board of Trustees of the
BIBLIOGRAPHY University of Illinois. The iperf
[1] Andrew S. Tanenbaum. Computer Networks, website.
Fourth Edition. Prentice Hall, 2003. http://dast.nlanr.net/Projects/Iperf/,
[2] Kevin Hamilton Kennedy Clark. Cisco May 2005.
LAN Switching (CCIE Professional De- [19] The ttcp website.
velopment). Cisco Press, 1999. http://www.pcausa.com/Utilities/pcattcp.htm,
May 2005.
ISSN: 2395-1303 http://www.ijetjournal.org Page 138
International Journal of Engineering and Techniques - Volume 3 Issue 3, May- June 2017