Opnet Atm

1
TCP/IP Over ATM Using ABR and UBR Traffic

Daniel Copeland, EC 4850, September 21, 2001
if one of the cells is damaged or lost depends on

Abstract- ABR and UBR traffic are analyzed whether ABR or UBR service is used.
in a TCP/IP over ATM network. The
network consists of two subnets connected by TCP Segment
an ATM cloud. Background traffic in the
network consists of real-time VBR traffic.
Foreground traffic consists of FTP clients IP Packet
requesting files from an FTP server in the
opposite subnet. Link Utilization, CS PDU
throughput, queuing delay, and FTP response
time are all analyzed for ABR and UBR
ATM Cell
traffic. It is determined that in a TCP/IP over
ATM network with relatively light traffic, ATM Cell
ABR and UBR perform identically.
ATM Cell
1. Introduction Figure 1. TCP/IP over ATM with AAL5.

This project consists of an OPNET
simulation comparing Available Bit Rate (ABR) 3. ABR / UBR
and Unspecified Bit Rate (UBR) ATM traffic in ABR traffic is used for delay sensitive
a TCP/IP over ATM AAL5 network. Real time applications and provides its own congestion
VBR video traffic is used as background traffic. information to the user. The user is then
Link utilization, throughput, and queuing delay expected to increase or decrease its traffic load
are all compared to determine the strengths and based on the congestion information provided.
weaknesses of ABR and UBR traffic. Since TCP also provides congestion control
Before discussing the simulation, a brief using window-based flow, there are two
overview of TCP/IP over ATM and the ABR and congestion control algorithms operating
UBR services is given. simultaneously.
TCP/IP conducts window-based
2. TCP/IP Over ATM congestion control using two methods. At first,
AAL5 is used to convert TCP/IP “slow start” is used to control the speed at which
packets into ATM cells for transfer over an ATM packets are sent. A congestion window is kept
network. Since TCP/IP packets are of varying which opens every time an ACK is received and
size and typically much larger than the 48 bytes closes every time a new packet is sent. Initially,
of payload in an ATM cell, AAL5 segments the the congestion window is only one segment long,
TCP/IP packets. This segmentation can have however, each time an ACK is received, the
consequences on TCP operation if portions of congestion window is doubled. This results in
the IP Packet are lost or discarded, and in some an exponential increase in the number of packets
cases severely affect data throughput. sent before an ACK is received [1].
Segmentation is accomplished as shown Once packets begin to be lost, the TCP
in figure 1. The IP packet is encapsulated in the congestion avoidance algorithm takes over.
convergence sublayer packet data unit (CS After reducing the window to half the size it was
PDU). The CS layer then adds an eight-byte when packets began to be lost by slow start,
footer and pads the PDU so that it is a multiple congestion avoidance increases the window size
of 48 bytes. This CS PDU is then split into 48 by one each time (as opposed to slow starts
byte ATM cells, each of which has a five-byte exponential increase). These algorithms allow
header. The ATM cells are sent normally over the sender to control packet flow based on
the ATM network and at the distant end the CS network conditions [1].
layer reassembles the IP packet. What happens
2
ABR uses a different method of flow Cisco 3660 router with a second OC3 line, and to
control. Cells are first sent at the Initial Cell an ATM workstation with an OC1 line. The
Rate (ICR), which is negotiated with the Cisco 3660 router uses ATM to communicate
network. The source the increases the data rate with the ATM switch, and a 10BaseT Ethernet
to the Allowed Cell Rate (ACR) which must be line to communicate with the Ethernet Hub.
less than the Peak Cell Rate (PCR) negotiated Finally, an FTP Server is connected to the hub.
between the sender and receiver. Resource cells The FTP server is used for FTP transfers to the
are sent separately from the data cells to indicate clients in subnet 2. The ATM workstation is
necessary changes in the cell rates. These used to generate VBR-RT background traffic
resource cells travel both to and from the sender, with the ATM workstation in subnet 2.
and may be updated by switches along the way if
they detect congestion in the network [2].
These two congestion control methods
act on top of each other when TCP traffic is sent
over an ATM network using ABR. In [2] it is
shown that TCP can achieve its maximum
throughput over ATM using ABR if the buffer
sizes of ATM switches are large enough to not
adversely affect the TCP congestion algorithms.
UBR traffic does not provide
congestion control and UBR cells that cause
network congestion to become too great are
discarded. This means higher layer protocols
such as TCP that use UBR must ensure on their
own that their traffic reaches its destination.
In [3] it is shown that TCP alone
exhibits a throughput of over 90% while
throughput for TCP over ATM using UBR has
throughputs as low as 34%. This demonstrates
the severe reduction in performance that can
occur when a ATM cell containing a portion of a
TCP segment is discarded. Figure 3. Subnet 1.
4. Simulation Network
The network used for the OPNET
simulation, shown in Figure 2, consists of two
subnets connected by an ATM cloud. The
subnets are connected to the cloud with OC3
lines that each has a background utilization of
95.96%. This leaves approximately 6Mbps of
bandwidth for the subnets to use.
Figure 2. Network Backbone.
Subnet 1, shown in Figure 3, consists of

Figure 4. Subnet 2.
a ForeSystems ASX-200BX ATM switch
connected to the cloud with an OC3 line, to a
3
Subnet 2, shown in figure 4, is very and FTP traffic together never load the network
similar to subnet 1. The only difference is that in down sufficiently to affect each other.
place of the FTP server are three FTP clients
connected to the hub with 10BaseT lines. All
other components and lines are identical to
subnet 1.
5. Simulation Traffic
The ATM workstations communicated
with each other using VBR-RT video at 30
frames/sec, 352 x 240 pixel per frame. Video
sessions lasted 60 seconds each, with an average
of 60 seconds between calls. The FTP clients
each conducted downloads of 100KB files, with
an average of 60 seconds between downloads.
Some downloads were conducted while video
ATM traffic was being passed, and some while
no video traffic was passed.
6. Results
One simulation was run with the
TCP/IP over ATM traffic set to ABR and Figure 5. ATM backbone Utilization,
another simulation run with the traffic set to Throughput, and Delay.
UBR. Unfortunately the traffic used was unable
to sufficiently load down the network such that
differences in the ABR and UBR traffic could be
seen. Thirty-minute simulation time simulations
(taking over three hours real time) were
originally run, but showed no difference in the
results between ABR and UBR. Ten-minute
simulation time simulations were used with for
the graphs shown below. Since the ABR and
UBR simulation resulted in identical output, only
the results of the UBR simulation are shown.
Figure 5 shows the utilization,
throughput and queuing delay for the link
between subnet 1 and the ATM cloud.
Utilization and throughput are fairly steady at
about 96% due to the background traffic on the
link described earlier. Queuing delay is also
fairly steady, never rising above 60
microseconds. The FTP and video traffic used
was simply not enough to load down the network
at any time. Figure 6. Ethernet Hub to Router Link
Figure 6 shows the link statistics for the Utilization, Throughput, and Delay.
Ethernet link between the router and hub on
subnet two. The 10BASET network is never
even marginally utilized due to the low file
transfer sizes of the FTP clients.
Figure 7 shows the client data for the
simulation. The important thing to notice is that
file transfers taking place at the same time as
video conferencing appear no different than
when no video traffic is being passed. The video
4
[2] S. Kalyanaraman, R. Jain, S. Fahmy, R.

Goyal, F. Lu, and S. Srinidhi, “Performance
of TCP/IP over ABR service on ATM
Networks,” IEEE, May 1996.
[3] A. Romanow and S. Floyd, “Dynamics of

TCP Traffic over ATM Networks,” IEEE
Journal on Selected Areas in
Communications, vol 13, no. 4, pp. 633-641,
May 1995.
Figure 7. FTP Client and ATM Video Results.
7. Conclusion
The network used in this simulation was
too small to effectively compare the differences
between ABR and UBR in a TCP/IP over ATM
environment. The thirty-minute simulation took
over three hours to run and several runs were
required to properly configure the network. It is
estimated that a simulation which could properly
load the network down to test ABR vs. UBR
traffic would require nearly a day to run thirty
minutes of simulation time with the resources
available. Unfortunately, this is beyond the
scope of this project.
What this project did show is that even
a highly utilized ATM network is easily able to
handle moderate traffic from a relatively small
number of hosts. Despite the fact that at times
all three FTP clients were downloading large
files and the ATM workstations were conducting
teleconferencing, no degradation in the network
was ever observed.
8. References
[1] W. R. Stevens, TCP/IP Illustrated, Volume
1, Reading: Addison-Wesley, 1994.

Opnet Atm

Uploaded by

Copyright:

Available Formats

Opnet Atm

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Opnet Atm

Uploaded by

Copyright:

Available Formats

1

TCP/IP Over ATM Using ABR and UBR Traffic

if one of the cells is damaged or lost depends on

1. Introduction Figure 1. TCP/IP over ATM with AAL5.

Figure 2. Network Backbone.

Subnet 1, shown in Figure 3, consists of

[2] S. Kalyanaraman, R. Jain, S. Fahmy, R.

[3] A. Romanow and S. Floyd, “Dynamics of

Figure 7. FTP Client and ATM Video Results.

You might also like