An analytical model for evaluating utilization of tcp reno

Transcription

1 An analytical model for evaluating utilization of tcp reno mohammad mehdi hassani reza berangi Abstract: This paper presents an analytical model for TCP Reno. For this model an algorithm is derived to calculate the utilization and packet drop rate. The accuracy of the model is verified by comparing the calculated results versus simulation results. These results show that the TCP Reno is superior to another vesion(tahoe) by having higher percentage of utilization and lower percentage of packet dropping rate. Key words: tcp reno,utilization,markovian model.congestion control,fast recovery INTRODUCTION TCP is known to be the most useful transport layer protocol all around the globe [, ]. It provides a safe and connection oriented service on the network. Varieties of applications on the internet depend on the TCP efficiency and therefore, the analysis of the TCP efficiency is important for conjecture of this usage. Network congestion is one of the factors that reduces the TCP efficiency particularly, when variety of traffics are on the network. Certain input parameters, such as: the number of users, different network applications, network capacity and etc. are used to control the congestion. As the network bandwidth increases [3] and various network applications are created, more attention is paid on the congestion control mechanisms by focusing on the flow control management on the TCP. The main objective of this paper is to analyze the existing congestion control algorithms by Markov model and compare their utilization and throughput. Initially, the existing TCP congestion control algorithms are explained. After this, TCP Reno [5] which employ this algorithm is explained. Finally, this protocol is explained with Markov model and an algorithm is obtained for each protocol to calculate their utilization. The calculation results are verified by simulation and this TCP protocol is assessed based on these results. congestion control Reno mechanism is another generation of TCP which has two other mechanism in addition to with Tahoe. it has fast recovery and fast retransmit. In this part we analyze the main control compression algorithm totally and explain their function. for example:. Slow-Start. Congestion avoidance. Fast retransmit. Fast Recovery SLOW START Algorithm create for solve the compression problem function of this algorithm base on the CWND when we have new connection, CWND take digit one as default value. if receiver, receive a packet with sequence number (n),. with send ACK packet (acknowledge) verification receive packet number (n) ACK packet contain information about sequence number of other packet (n+). in fact ACK packet specify sequence number of next packet that sender would send. when receiver, receive ACK packet, addition digit one to CWND [4,]. CWND increase exponentially, CWND limited on network capacity, this limitation define bye the least value of sender and receiver's frame size. figure show TCP make connection between receiver and sender. at first size of frame is (one packet send ). - IIIB.4- -

2 After send of packet, sender wait for acknowledge at next step when sender receive acknowledge value of compression window become and two packets will send. if both packets receive correctly value of compression window become four. we can guess this growthing is not exactly exponentially. because, perhaps the receiver instead of two receive send one ACK. congestion avoidance This algorithm used for solving the problem of lost packet. Traffic occur when rate of receive packet sent. [5] normally to operation show lost packets :. Time Out. Receiver, Receive the same acknowledge Figure. Increase size of CWND [5] Congestion avoidance and Slow-Start are different control algorithm (3) but their function dependent on each other. the control mechanism that made by mixing these two algorithm use two parameter for specify number of send packet.. CWND. Slow-Start threshold Size (SSTHRESH) the number of sending packet by sender obtain from relation () and receive window, is the size of buffer in receiver that send to the sender. () Min(Receive Window, CWND). when we connect, TCP put the packet's value in CWND then slow-start algorithm will start. when compression occur half of current window store in SSTHRESH and CWND assign with one packet again. TCP run slow-start algorithm at start of connection and its continue until the size of CWND I smaller than SSTHRESH. after that TCP run congestion avoidance algorithm. and in this algorithm value of CWND increase linear. and this increase obtain from relation (). () segmentsize*segmentsize/cwnd in relation () size of segment equal with the number of sender packet. fast retransmit - IIIB.4- -

3 Last model of TCP recognize, losted packet and network compression by timeout mechanism. when the packet sent, receiver wait for period of time (RTO). If receiver, receive the packet correctly, before the end of RTO time send ACK for sender. if one packet receive out of sequence, receiver send the simulate ACK (dupack). Dupack can occur in effect with two operation losting packet and receiving packet out of sequence. in this state may be one or two dupack send before second sending on base of sequence number. if we have more than to dupack, it means that packets lost. as we see in figure TCP understand losting packet with dupack and send a lost packet again. Figur. Fast retransmit algorithm in packet loss. Fast recovery When receiver, receive three packets we recognize that packet losted, so the value of CWND become equal with half of current windows size and obtain the value of SSTHRESH from relation (3) and send a packet again. the size of CWND obtain from relation (4) and when sender packet's receive SSTHRESH value put in CWND again.this algorithm performance like slow-start on facing to timeout problem. SSTHRESH=Min (CWND old, Receiver's advertised window)/(at least MSS) 3 CWND new =SSTHRESH + number of dupacks 4 Figure 3 show the function of fast recovery and fast retransmit. in last years various generation of this control algorithm appear, so we compare them. - IIIB.4-3 -

4 Figure 3.fast recovery algorithm Tcp reno This model use all of the TCP Tahoe equipment. the difference between TCP Reno and Tahoe, is using of the fast recovery algorithm [6]. it means if timeout appear CWND assign with. when receiver, receive 3 dupack, fast recovery algorithm called. Peresent Analyze Modeling In this part we explain TCP efficiency in transmit between source and destination for Reno, with Marcovian model. image the state that the mobile receiver and mobile sender in wireless network is uploading and downloading information from workstation. With Marcovian model we conjecture sender upstream by sender each state with marco model show size of sending frame, state increase exponentially so, PI show the state which TCP window size equal with i Analyzing TCP Reno Figure 4 show Marcov chain for TCP Reno. on base of Reno model three various event appear :. in each state ACK or all packet or all packet the value of windows size equal with i and we enter next state.. in each state if three dupack receive, CWND should equal with half of last value. And in Marco model we do it by return to last state and CWND equal with i-.. if timeout occur then CWND equal with and Slow-Start algorithm start from begin, in Marco model from current state reach two state 0. - IIIB.4-4 -

5 Figure 4. Marcovian chain TCP reno An algorithm can be derived based on this model to calculate the efficiency of the Reno. Similarly, an iterative algorithm is derived based on this model to calculate the efficiency of this TCP variation. The algorithm shown in (5) uses (4) to solve a system of n + equations with n variables by assuming k= n. pn = p n n+ n + n p n = p n _ for( i = n to 0) n (5) l = ( + i l pi = } i i+ ) p i+ + ( i The utilization and drop rate that are calculated with these algorithms are shown in Figures 6 and 7. Simulation study In order to verify our analytical model, the utilization of the TCP Reno is obtained by computer simulation. A scenario is simulated using glomosim with a network of two subnets (receiver and sender) that communicate through a server with a base station. The base station buffer size is assumed to be 00 packets. The simulation runs 0 times, each lasting 00 seconds. Figure 6, shows both simulated and calculated utilizations from these two algorithms versus window size. As can be seen, the simulation results closely follow the calculation results as expected. The drop rate versus window size is also found for this algorithm and the results are shown in Figure 7. + ) p i+ - IIIB.4-5 -

6 utilization* simulation reno utilization computed reno utilization window size Figure 6.tcp tahoe and reno utilization droprate* simulated reno drop rate computed reno droprate window size CONCLUSIONS Figure 7.tcp tahoe and reno drop rate In this paper TCP Reno is explained based on its Markov model. this model is used to derive a iterative algorithm to calculate utilization and drop rate. The calculation results are verified by simulation results. It is shown that TCP Reno outperforms other version of TCP (Tahoe) because of using fast recovery algorithm. REFERENCES [] Stevens, W., "TCP Slow Start, Congestion Avoidance, Fast Retransmit,Recovery Algorithms," RFC 00, 999, [] Padhye, J., Floyd, S., On Inferring TCP Behavior, Computer Communications eview ACM-SIGCOMM,Vol. 3, August 00. [3] Ewerlid, A., Reliable communication over wireless links, in NordicNRS, Sweden, Apr IIIB.4-6 -

7 [4] Jacobson, V., "Congestion Avoidance and Control," Computer Communication review, 8(4), August 988, pp [5] Kirov, M.G., "A Simulation analysis of the TCP control algorithm," International Conference on computer system and technologies, CompSysTech 005. [6] Fall, K., Floyd, S., Simulation Based Comparisons of Tahoe, Reno, TCP. Computer communication review 6(3),July 996, pp.5- [7] Jacobson, V., "Congestion avoidance and control," in Proceedings of 88 workshop, ACM SIGCOMM, ACM. press, Stanford, CA, 988 PP ABOUT THE AUTHOR Mohammad mehdi hassani, student of major master of science,iran university of science & technology, Reza berangi, PhD, Department of Computer Systems,iran University of science & technology,, -mail: - IIIB.4-7 -