Improving Transport Layer Performance - Assignment Example

IMPROVING TRANSPORT LAYER PERFORMANCE (Using Novel Medium Access Control and Fast Collision Resolution Algorithm) INTRODUCTION In computer networking, the TRANSPORT LAYER provides point-to-point communication services between different hosts. The popular and most common protocol such as TCP/IP employs at least one of these transport layer protocols. These are Transmission Control Protocol (TCP) and the Datagram Protocol (UDP). Using the services of the network layer beneath, it provides transparent data transfers that are generally useful in various methods of flow control such as data integrity verification, effective error recovery, and other multiplexing and de-multiplexing sessions. TCP AS A TRANSPORT LAYER PROTOCOL TCP is a dependable and reliable transport protocol performing connection oriented data transfers with flow and congestion control. It re-transmits the data segment whenever data segment and acknowledgements are lost due to packet damage or network congestion in a predetermined timeout interval. The performance of these timeouts and re-transmissions of data segments is the main issue of the research. TCP, when tasks to deliver large amount of data to the other end invoked the procedure called "Slow Start". This procedure controls that amount of data to transmit between host (sender and receiver). It monitors the rate of new packets delivered into the network and at the same time monitors the rate of acknowledgements returned from the receiver. The "Sliding Window Mechanism" where the slow start procedure depends, permits the sender to transmit multiple packets successively then wait for the receiver's acknowledgement. If the connection is successful, the receiver will now send it's acknowledgement along with the information called "receiving window size" that indicates the limited amount of data segment a sender can transmit. The sender is now restricted to send the amount of data segment specified in the receiving window size. On the other hand, the TCP sender also manages its data transfers using "congestion window" which is set to one segment higher whenever an acknowledgement is received. This indicates that sender can only send restricted amount of data segments specified in both "window". In situations where packets are lost because of damages in transit or network congestion, TCP employs "congestion control" algorithms. It operates with "slow start" procedure by maintaining the "congestion window" and the slow start threshold size. If the data segments sent are not acknowledge in the allotted retransmission period or time out (RTO), TCP performs retransmission procedures, setting "congestion window size" to one segment higher and "slow start" threshold size to one-half of the current window. When the receiver successfully acknowledges the "re-transmitted" data segment, TCP invokes either "slow start" or "congestion avoidance". It only uses the 'slow start' procedure and increases the congestion exponentially when the congestion window is equal or less than the size of the slow start threshold. Otherwise, 'congestion avoidance' is triggered increasing the congestion window by half of the current size and continue until both congestion window and threshold size is equal. This 'congestion avoidance phase' when using 'slow start' degrades the data throughput severely and can be solve by re-designing the procedures to speed-up the recovery of the connection. The designs of Fast retransmit and Fast recovery detects a segment loss by constantly monitoring duplicate 'acknowledgements'. It works by consecutively sending 'sequence numbers' corresponding to the lost segment. Three or more duplicate acknowledgements indicate loss segments and therefore "TCP fast retransmit mechanism carries out retransmission of the missing segment even before the retransmission timers expires" (Kwon et. al 2002). The designs maintains that there is no need to use 'slow start procedure' if there is still normal data flow between the sender and the receiver because it reduces the transmission rate hastily. In addition, fast recovery mechanism performs congestion avoidance after missing segments fast retransmit. THE MAC (Medium Access Control) ALGORITHM The MAC protocol is a popular contention-based medium. It is a carrier sense multiple access/collision avoidance algorithm used with IEEE 802.11 LAN. In general, packet transmission cycles are considered successful when transmitted and acknowledged at the other end of the line. When a station wants to transmit a packet, MAC uses its carrier sensing mechanism to check the medium status. If it is idle, the transmission begins otherwise the station will defer until such time that the medium is available. After the idle time (Distributed Interframe Function Space, DIFS), the station will now perform a 'backoff' procedure where it will set the backoff timer to a random backoff time based on the existing contention window (CW) size. The carrier sensing mechanism will continually determine whether an activity is occurring during each backoff slot. If no activity is detected, the backoff procedure will decrement its backoff time by a slot time. (BTnew=BTold - aSlotTime). The same procedure applies when the medium is busy; the backoff procedure will be suspended for a while until the medium becomes idle after the DIFS. In short, transmission begins whenever the backoff timer is equal to zero and when the station transmits a packet and receives and acknowledgement with no errors within a short inter-frame space (SIFS), the transmission is considered successful and contention window (CW) will be reset to its minimum size. Otherwise, CW will be increase up to its maximum value. The minimum value for minCW = 31 and the maxCW= 1023 as per IEEE 802.11 standard. This is a binary exponential backoff intended to resolve collisions. The major factors to consider in the degradation of throughput performance in the IEEE 802.11 MAC protocol under high traffic loads are: a) transmission failures due to packet collisions b) backoff at each contention cycle resulting in more idle slots. If we are going to assume that, all stations have packets to transmit during the peak period. The following expression represents the throughput of one transmission cycle. Where: E[Nc] represents the average number of collisions. E[Bc] is the average number of idle slots resulting from backoff. ts is the length of a slot (aSlotTime) is the average packet length From this scenario, we can now derive the best possible maximum throughput for an effective packet transmission whether the number of collision is zero or the number of idle slots is also equal to null. It would be best to divide the average packet length to the combined values of the total packet length, inter-frame space, acknowledgement data limit, and the distributed inter-frame function space. This is possible using the probability of packet transmission ptrans(i) in each stations contention period. Assuming that the backoff timer is chosen randomly for each station, the current contention period would be entirely depending on the backoff timer. Where Bi is the backoff timer of station i. If the backoff time is zero then station (i) will transmit a packet immediately. This can be converted into : This is the basis of contention-based MAC algorithm. It assigns a backoff timer of zero for a particular station and assigns a value of one (1) to all other stations to achieve perfect packet scheduling producing maximum throughput. Although this scenario is not a standard practice, this provides a basic idea to design an efficient contention-based MAC algorithm. COMBINING MAC PROTOCOL WITH FCR The designs of MAC protocol have these following operation characteristics: a. Small random backoff timer for each station transmitted a packet successfully that would decrease the idle slots for each contention period. b. Larger random backoff timer for stations that deferred their packet transmission during the contention period. (All stations with non-zero backoff timer) c. Rapid change of state for random backoff timers. State change from large to small. d. Adapting a Fast Collision Resolution Algorithm (FCR). To solve the major deficiency of IEEE 802.11 based MAC Protocol. FCR is using smaller contention window size minCW and larger maximum contention window size maxCW. Automatically increase the overall contention window size when a station is in both collision and deferring state. It also reduces the backoff timers exponentially fast when consecutive idle slots are detected. THE MAC & FCR PERFORMANCE TEST Using a direct sequence spread spectrum (DSS) for simulation with a parameter value equal to the following: Transmission rate for data and ACK frame are 11Mbps and 2Mbps each. SIFS=10sec DIFS=50sec A slot time of 20sec aPreambleLength of 144 bits aPLCPHeaderLength equal to 48 bits Bit Rate of 2, 11 Mbps. The average transmission time for a single packet is given by where ts= slot time. During the test, FCR was assigned the maximum successive packet transmission limit of 10 in 100 second simulation times. The throughput results show the IEEE 802.11 MAC algorithm's very poor performance as the number of active stations increases. This is due to high collision probability as the quantity of stations becomes bigger. In FCR, all stations increased their contention window size except for one with successful packet transmission. This is an indication that all stations can swiftly acquire the proper contention window size to spoil and reduce the probability of future collisions. On the other hand, a station with successful packet transmission has lesser CW size with a value of three (3) compared to IEEE 802.11 MAC algorithm with minCW of 31 therefore reducing medium idle time. FCR algorithms efficient collision resolution strategy notably improved the throughput performance over the IEEE 802.11 MAC algorithm and did not degrade much as the number of stations increases. The FCR algorithm indeed achieved high throughput performance while maintaining the straightforwardness of implementation in LAN and Wireless Networks. This is simply by intelligently changing the contention window size and redistributing the backoff timers in all active stations for both successful packet transmission and collision. Each station resolving collision quickly and in coordination can greatly enhance throughput performance by reducing idle slots time. This approach is well suited to high traffic layers like TCP and UDP. OTHER RELATED RESEARCH AND IMPROVEMENTS Some other researches using different approach to improve the transport layer performance were done in various universities such Princeton, UCSC, and the University of Texas. The University of California conducted an older study back in 1999 to improve the transport layer performance through the LINK LAYER. This is the Transport Unaware Link Improvement Protocol or TULIP. It is probably the inspiration behind the later research because it also provides MAC acceleration features that are applicable to collision-avoidance. This is like the latest improvement in MAC with FCR using backoff timers as indicators. The difference is that "TULIP timers rely only on the maximum propagation delay over the link, rather than a round trip estimate of the channel delay" (Parsa et al 1999). The Princeton's researchers approach is to use REDUNDANT PATHS, which they believed, are common between pair of hosts based on a recent study conducted over the internet and overlay networks. The researcher's idea is to have a multi-path TCP or mTCP where stations can continue sending packets to an alternate path if the current path fails. mTCP cannot only reach higher end-to-end throughput but also become more reliable under path failure conditions. Note that the FCR approach is contention based and is using the same conventional path to resolve collision. mTCP can also alleviate aggressiveness problems by monitoring, detecting and suppressing paths with shared congestion. The researcher claims that mTCP reliability is often achieve by sufficiently disjointing paths to the destinations of interest. "Users of wireless networks can often have access to multiple communication channels simultaneously" (Zhang et al, n.d.). Probably the latest and more sophisticated approach is the Improvement of Transport Layer through Multi-homing using SCTP (Stream Control Transmission Protocol) done by the Department of Computer Science of the University of Texas in 2004. It refers to the concept of a single machine with multiple interfaces connected to different ISP simultaneously. SCTP unlike the traditional TCP is message base that allow a single endpoint to be associated with multiple IP. However, there maybe a number of extreme studies conducted to improve the transport layer performance but none of them possess the practicality and simplicity of the MAC with FCR approach done by the University of Florida in 2002. The good thing about this improvement is the fact that it maintains solving the contention and collision problem of the network in the congested path itself rather than redirecting packets in other path or creates multiple interfaces. It is certainly more feasible and economical to adapt the MAC with FCR approach if the improvement of the transport layer is really our main concern. References: Baji, Bhat, 2004, "Improving Throughput at the Transport Layer through Multihoming using SCTP", Department of Computer Sciences, University of Texas at Austin, , 07/19/2006, http://www.cs.utexas.edu/vishwas/documents/CMT.pdf Kwon, Fang, Latchman, 2002, "Improving the Transport Layer Performance by Using a Novel Medium Access Control Protocol with Fast Collision Resolution in Wireless LANs", Department of Electrical and Computer Engineering, University of Florida, , 07/19/2006, http:// www.city.londonmet.ac.uk/mcampbel/qc308/articles-late.htm Parsa, Garcia, Aceves, 1999, "Improving TCP Performance over Wireless Networks at the Link Layer", Computer Engineering Department, Baskin School of Engineering, University of California, < online>, 07/20/2006, http://www.cse.ucsc.edu/research/ccrg/publications/tulip1.pdf Zhang, Lai, Krishnamurthy, Peterson, Wang, n.d., "A Transport Layer Approach for Improving End-to-End Performance and Robustness Using Redundant Paths", , 07/20/2006, http://www.cs.princeton.edu/rywang/papers/usenix04/ Read More