INTRODUCTION
TCP Reno, New Reno and SACK TCP are all standard TCP congestion algorithms. Standard TCP Congestion algorithm selects the congestion avoidance phase which is known as Additive Increase/Multiplicative Decrease (AIMD) algorithm [1]. It means that when window size is increase given the maximum segment unit passes and therefore MTU is fixed for its transmission, so Additive increase occurs. When the sender window size is decreased then the congestion window on the receiver end is increased. Here, the estimate concerning the probability is calculated that the transmission of MSU has encountered an error. Under the algorithm of AIMD, the sender inquires the network looking for extra bandwidth by increasing their send rates till the congestion is detected. In low speed networks, TCP can exercise congestion control effectively. So if TCP standard congestion algorithm and AIMD is thoroughly studied, it is found that AIMD algorithm is very conservative and cannot be practiced for large window transmissions. Because for a large window, it is difficult for the user with large window size to recover after back-off algorithm is run and effective utilization of bandwidth cannot be ensured. In comparison TCP Illinois operates in a high speed networks. It takes the information of packet loss to determine the increment and decrement of window size. It attains highest throughput and allocates the network resources fairly and in a most innovative way. It combines with the standard TCP congestion algorithms to enhance the collective network performance.
REVIEW OF THE STATE OF ART:
Several Protocols have been introduced in the past to replace the TCP Standard congestion algorithms that apply in the highest speed networks. Comparing these protocols and giving an insight into the development of an effective and viable protocol, basic requirements have been laid down just to give a tip how a new protocol should live up to best of its performance [2].
Intra-Protocol Requirements: The requirements that the protocol should fulfill is the following mentioned below:
Incentive: If the users switch from the standard TCP to new protocol, they should have a large average throughput.
Shao Liu, Tamer Basar and R. Srikant have stressed in their research journal that TCP Illinois has so far achieved the better results in congestion avoidance against other standard TCP protocols. In data communication, while applying TCP Illinois against the standard, the Sender Window is kept high when no congestion is imminent. But when the congestion is imminent the window size is decreased. So the argument is the packet losses can be brought down if the receiver window size is also decreased. In this scenario the average queuing delay, is calculated to give directions when the congestion is imminent, and herein the queuing delay is the delay occurred when a packet sits in a routing queue (buffer) and about to be processed. TCP-Illinois uses the functions of fast recovery and fast retransmission acquired from New Reno Standard TCP Protocol. TCP Illinois has adopted the idea of selective acknowledgement from SACK TCP and avoid heavy congestion in an effective manner.
IMPLEMENTATION OF TCP ILLINOIS IN LINUX
Laboratory test is performed to tune TCP through congestion algorithm of TCP Illinois. Initially the configurations are performed using the terminal console application. Later Wireshark is used to analyze the network the performance of TCP Protocol. A dump file is downloaded to evolve a set of test cases to thoroughly check the network speed as well as congestion and enable the user to evaluate the difference of performances. Before Tuning and After Tuning comparison is given below to show the following graphs, like RTT Graph, Throughput Graph, IO Graph, Time/ Sequence (Stevens) and Time/Sequence (Tcp Trace). Slow Start is shown in before and after Tuning graphs of both Time/Sequence Graphs.
Before Tuning:
Buffer Size (Rmem)
Minimum: 4096
Default: 87380
Maximum: 2043904
Window Buffer Size (Wmem)
Minimum: 4096
Default: 16384
Maximum: 2043904
Window Size: 6432
Link Delay: 1151. 648216 packets/seconds.
Bandwidth: 9.048 MBit/Seconds
After Tuning
Buffer Size (Rmem)
Minimum: 8192
Default: 87380
Maximum: 2043904
Window Buffer Size (Wmem)
Minimum: 16384
Default: 16384
Maximum: 2043904
Window Size: 6432
Link Delay: 1195.382883 packets/seconds.
Bandwidth: 9.378 MBit/Seconds
Conclusion
In conclusion it can be laid down here that after TCP Tuning through TCP Illinois algorithm, throughput, Round Trip Time, Graphs of Stevens and Graph of TCP Trace showing Time/Sequence, where slow start is shown have been optimized. Bandwidth and transfer of packets have shown a marked increase. But Link delay has been increased as well. Because, Round Trip Time is normally over estimated by TCP Illinois. Errors occur in maximum RTT estimation as Linux allocates the time slot for the packets at the TCP layer instead of allowing their arrival at the Network Interface Card. Therefore, delays in Processing were also applied to allow the unacknowledged packets to be processed within due time.
References
