Performance Evaluation of
TCP over ATM in Broadband Satellite Networks
By - Ajit Gundale
__________________________________________________________________________________________________________________________
TCP is reliable, connection-oriented, end-to-end error, flow control and congestion control protocol. It is widely used protocol for data transfer in packet form in the Internet. It is designed to work independently of lower layer, such as ATM, to transfer data. Its basic implementation is unsuitable for high speed and long delay networks and therefore modifications are suggested to enhance its performance. The proposed extensions are use of large window size in order to accommodate high BDP value of high speed & long delay network and TCP time stamp for precise estimation of round trip time (RTT). Various TCP schemes are suggested to improve TCP behavior on high BDP networks by slow start phase at the beginning or retransmission policy to have multiple lost segment repetition in one round trip time.
ATM is a scalable cell switching and multiplexing technology for Broadband Integrated Services Digital Network (B-ISDN). It is widely implemented in LANs and WANs. The TCP over ATM provides maximized performance when transmitter and receiver both are synchronized properly and congestionless situation on the channel. The performance of TCP over ATM is characterized by a metric throughput.
The
low throughput of TCP in ATM networks is mainly due to the reason- The loss of
any cell in a packet corrupts the corresponding whole packet, TCP retransmits
this corrupted packet, the packet generates a bundle of cells, this process is
repeated and finally generates vast retransmission duplicates. The other reason
is that TCP should wait for time-out period, which is greater than Round Trip
Time (RTT) and is usually too big, in order to retransmit the lost
packet when a cell (or a packet) is dropped during its transmission. And if it
is a long distance communication then actual throughput is hardly to get.
TCP Application sends data in large packets. These
packets must be segmented to fit into the ATM cell structure before being put
on the network. The effective throughput of TCP over ATM networks may
be quite lower than that over packet-based networks, when ATM cells are dropped
at congested ATM switches. Most cell losses occur at congested switches in ATM
networks while physical links usually have very low bit error rate (BER).
When TCP-ATM application is used in satellite
networks, congestion may occur in ATM switches. Thus ATM Congestion Control
mechanism is necessary to have maximum throughput. Due to unavailability of
required bandwidth or output queue size, cells are dropped. By selecting proper
TCP parameters and output queue size, % cell drop can be reduced giving
improved throughput. Hence it is important to study congestion control of TCP
over ATM in satellite network. ATM rate-based flow control mechanism
effectively reduces cell loss and buffer requirement in ATM switches, and helps
to enhances overall performance in ATM networks.
Following are the observations during studies of
performance of TCP over ATM in satellite networks.
Optimization of TCP parameters
in ATM networks with
Background Traffic applied
This simulation is done to
analyze the performance of a TCP over ATM in satellite network. The analysis
consists of optimization of TCP parameter such as Throughput, Retransmission
Percentage, and Maximum Segment Size.
Network shown below is selected to find various TCP
parameters:
Network Parameter Detail:
TCP Application Parameters:
Bit
Rate |
155 Mbps |
Buffer
Size |
Variable |
Max.
Segment Size |
Variable |
Buffer Management Algorithm |
EPD |
BTE parameters:
Max. Output queue Size |
Unlimited |
Max. Input queue Size |
Unlimited |
Rate Based Switch Parameter:
Delay to Process |
0 msec. |
Output Queue Size |
Unlimited |
ATM congestion control
algorithm |
Applied |
Selection of Maximum Segment
Size (MSS)
Here TCP Buffer Size is kept
constant and Max Segment Size is varied and effect is seen on Throughput and maximized throughput is
observed for certain MSS.
Constant Parameters |
Buffer size (bytes) |
9000 |
Variable Parameters |
Max. Segment Size (octets) |
512,5000,9180,
10000 |
Graph below shows Effect of
variation of Max. Segment Size on
Throughput
Here TCP Buffer Size in
network is varied and its effect is seen on Throughput
and Retransmission Percentage.
Constant Parameters |
Max. Segment Size (octets) |
9180 |
Variable Parameters |
Buffer size (bytes) |
7000,8000,9000, 32000 |
Graph below shows Effect of
variation of TCP Buffer size on
amount of percent
Retransmission
CONCLUSION:
ü Background traffic helps in proper utilization of the link. When traffic from TCP application is less and bandwidth is available this background traffic utilizes the remaining bandwidth.
ü From simulation it is observed that the Maximum Throughput for TCP connection with background traffic in Satellite based network is available with TCP buffer size of 9000 octets and Maximum Segment Size of 9180 octets.
ü For above optimized values minimum Retransmission is observed all TCP applications.
ü For Maximum Segment Size set to 9180 octets the Maximum Throughput is achieved, it concludes that MSS is approximately equals to TCP buffer size.
ü For background traffic, if the number of TCP connections is increased, the overall link utilization also improves.
Thus following parameters are assigned with optimized values in future simulation studies:
TCP Buffer Size |
9000 octets |
TCP MSS |
9180 octets |
Switch Slot rate |
155 Mbps |
Above parameter values ideal BTE and switches are assumed.
TCP over ATM in Satellite
Network
The following simulation put focus on optimization of B-TE switch parameters and effect of asymmetry on forward and reverse rate in satellite network. For simulation of both Generic and Rate based switches are considered.
Following network is assumed for simulation.
Parameters for TCP
Application:
l Bit Rate: 141 Mbps
l Buffer Size: 9000 bytes
l Mean Processing Time: 3 msec
l Max .Segment Size : 9180 octets
B-TE parameters:
l Max. Output queue Size 10
l Max. Input queue Size 10
Switch Parameter:
l Processing delay 0.3 msec.
l Output Queue Size kept variable
l Buffer Mgt. Algorithm Triggered
l Congestion Algorithm Applied
1.
(a) Generic
switch format Network
Effect of variation of Switch Output Queue Size.
Here Switch Output Queue Size of network is varied and its effect is seen on % Cell Drop of the Switch.
Constant Parameters |
Uplink and Downlink |
Link rate ( Mbits /sec) |
141 |
Variable Parameters |
Switch |
Output Queue size (cells) |
1,5,10,25,50 |
Graph
below shows effect of variation of Switch Output Queue size on Cell Drop.
Effect
of variation of Switch Output Queue Size.
Here
Switch Output Queue Size of network is varied and its effect on % Cell Drop of
the Switch is observed.
Constant Parameters |
Uplink and Downlink |
Link rate (Mbits/sec) |
141 |
Variable Parameters |
Switch |
Output Queue size (cells) |
0,5,10,50 |
Graph
below shows Effect of variation of Switch Output
Queue size on Cell Drop.
Above graph shows % Cell Drop in accordance with the Switch Output Queue Size. Here
the effect of Delay on Cell Drop is observed.
2.
(a) Performance of TCP over ATM in asymmetric
satellite network-
Constant Parameters |
Uplink |
Link Rate (Mbps) |
50 |
Downlink |
50 |
||
Variable Parameters |
Switch |
Output Queue size (cells) |
0, 5, 10, 50 |
Above graph shows variation
in the throughput for different switch output queue size
2.
(b) Performance of TCP over ATM in asymmetric
satellite network-
Constant Parameters |
Uplink |
Link Rate (Mbps) |
25 |
Downlink |
50 |
||
Variable Parameters |
Switch |
Output Queue size (cells) |
0,5,10,25,50 |
Above graph shows variation
in the throughput for different switch output queue size
2.
(c) Performance of TCP over ATM in asymmetric satellite network-
Constant Parameters |
Uplink |
Link Rate (Mbps) |
50 |
Downlink |
25 |
||
Variable Parameters |
Switch |
Output Queue size (cells) |
0,1,5,10,25,50 |
Above graph shows variation
in the throughput for different switch output queue size
CONCLUSION:
Satellite Based TCP Network
Simulation:
ü This simulation is done to study %Cell drop by varying Switch Output Queue Size. If the incoming cells exceed the limit of queue size or there is no bandwidth available on output link then Cells are dropped.
ü
It is found that the Minimum % Cell Drop is
achieved when Switch Output Queue Size is set at 50 Cells which is sufficient
enough to handle all the incoming data from TCP connections using ATM
Congestion Control mechanism for Rate based Switches. As Switch Output Queue Size doubles the Cell
Drop reduces approximately by 2%.
ü In asymmetric satellite networks, for given switch output queue size minimum cell loss at the side of switches is observed when uplink is larger than downlink. The variation of Switch Output Queue Size reduces % Cell drop by 1%. Thus we get Minimum Cell drop when Switch Output Queue Size is set at 50 cells.
Thus optimized results are observed when different parameters are assigned values given in the table
TCP Buffer Size |
9000 bytes |
Max. Segment Size (MSS) |
9180 0ctets |
Switch Output Queue Size |
50 cells |
Switch Slot rate |
155 Mbps |
Switch Delay to Process |
3 msec |
Switch ATM Congestion algorithm |
Triggered |