CHAPTER 6. SIMULATION STUDIES OF THE PERFORMANCE OF SCTP
6.2. VERTICAL HANDOVER WITH THE BASIC SCTP 118
In this section, different transport-layer protocols are used in the vertical handover scenarios. In this simulation, Random WayPoint (RWP) similar to many previous studies [74],[75] has used as mobility model. Random waypoint is a simple model that is easy to implement and analyse. In the RWP model, the nodes or mobile users, move along a zigzag path consisting of straight legs from one waypoint to the next [74]. Based on this mobility model, the goodput is measured for different transport-layer protocol and results are compared. SCTP is simulated without enabling the multi-homing feature. The objective of this simulation is to analyse the potential benefit of using SCTP in a combined wired and wireless scenario at the presence of mobility and error on the communication links.
6.2.1. Simulation Scenario
The scenario consisted of:
A single mobile node placed in a 670m by 670m rectangle, using the Random Waypoint mobility model and working as a sink client to collect FTP traffic from the corresponding node (see Figure 6-2)
Four base stations two of them belong to HA domain and the other two of them belong to FA domain in order to provide both inter-domain and intra-domain handovers. They are distributed as shown in Figure 6-3
and each base station is able to handle a transmission range of up to 250m (see Figure 6-3)
Four routers, with following functionalities: (See Figure 6-2)
♦ Internet Service Provider (ISP) router is an access router that connects the corresponding node to the public network.
♦ Gateway router is a backbone router that connects the mobile network to the public network or the Internet.
♦ Home Agent (HA) is part of the mobile network components which provides the facility to allocate an address to the MN and tunnels the packets towards the current position of the MN. ♦ Foreign Agent (FA) is part of the mobile network components
which belongs to a different domain of the HA. Its function is to provide a CoA for the MN, informs the allocated CoA to the HA and the CN and finally handle the traffic to/from the MN when the MN is located inside the coverage area of the FA. Correspondent Node, works as a server to generate FTP traffic. In the
simulation, different transport agents provide a reliable connection from this node to the MN.
All wired links have a bandwidth of 5 Mbps. A TCP connection (or an SCTP association) is used to transport the data generated by the FTP. The total simulation time is 200 seconds. Tahoe, New Reno and SACK enabled versions of TCP were used for the simulations. The MTU for each link was kept at 1500 bytes. The TCP segment size and SCTP data chunk were kept at 1000 bytes. The initial congestion window size for both TCP and SCTP were kept both equal to 2*MTU. The speed of the MN was a random value between 0 and 30m/s using a random waypoint mobility model. The base stations were distributed in a 670m square as shown in Figure 6-3 to have the maximum coverage in the region.
Figure 6-2: Simulation Topology in a wired-cum-wireless scenario
Figure 6-3: Distribution of BSs in a 670m*670m area
6.2.2. Packet arrivals Comparison
In order to analyse the connection robustness, a number of experiments have been done to compare the received data for different versions of TCP and SCTP when the MN has a RWP mobility model. Figure 6-4 shows the aggregation of the received data during the simulation time in two different scenarios. In the error-free environment all the TCP extensions delivered almost similar amount of data. TCP-Tahoe, TCP-NewReno and TCP-SACK have similar mechanisms for preventing congestion collapse and finding an appropriate rate of transmission
compared to various versions of TCP in the simulation as congestion control algorithms for both TCP and SCTP are similar. The difference in data rate between SCTP and TCP is due to the NS-2 implementation of TCP and SCTP.
Figure 6-4: Comparison of aggregation received data in zero drop and 5% loss rate scenarios
The main differences between these protocols are in congested and high error- rate scenarios that the packets are subjected to drop or loss. The results show TCP-SACK has better performance when a high error-rate was applied in the wireless path. The SACK-enabled segments provide the TCP sender with some extra information of the status of the destination’s receiving buffer. In SACK for every received packet, the receiver produces a reply which contains further information in the header of the segment in the form of an option. Hence, in the event of packet loss the sender can resend only the exact packets that have been lost in transit and avoid producing unnecessary retransmission. TCP-NewReno produces least performance as it primarily optimised to work with the burst error scenarios. SCTP shows better performance compare to TCP-SACK in the same situation. Original SCTP does not include a Fast Recovery mechanism, as found in NewReno-TCP and SACK-TCP and later TCP variants. As specified in RFC2960 [3], “because cwnd in SCTP indirectly bounds the number of outstanding TSN's, the effect of TCP Fast Recovery is achieved automatically with no adjustment to the congestion control window size”. This built-in fast recovery system along with the benefit of SACK algorithm implemented in SCTP makes this protocol robust in high error-rate scenarios.
12.518 12.434 12.464 13.463 7.424 7.036 7.788 8.707 0 2 4 6 8 10 12 14 16
TCP-Taho TCP-NewReno TCP-SACK SCTP
Transport Layer Protocol
R e ci ved D a ta ( M B ) No Packet lost 5% Packet lost
Figure 6-5 and Figure 6-6 show the aggregation of packet arrivals at the MN. The stationary parts on the curves indicate disconnectivity at that point, which are the impact of handovers at specific times.
Figure 6-5: Comparison of aggregation data-packet arrival in different transport layer protocol with handovers based on MIP in an error-free environment
Figure 6-6: Comparison of aggregation data-packet arrival in different transport layer protocol with handovers based on MIP and 5% uniform packet losses
The results presented in Figure 6-5 and Figure 6-6 show that the current SCTP implementation performs almost as well as TCP when there are no losses. However, SCTP seems to perform better in the presence of losses, as it benefits form built-in fast recovery system and does not enforce strictly ordered delivery.
0 2 4 6 8 10 12 14 16 0 50 100 150 200 Time (s) aggr ega ti on of dat a packet ar ri val ( M bps) Tahoe New Reno SCTP SACK 0 1 2 3 4 5 6 7 8 9 10 0 50 100 150 200 Time (s) agg reg at io n of d a ta pack et ar ri val ( M bp s) Tahoe New Reno SCTP SACK MB MB
TCP session. Therefore, TCP detects a gap in the received sequence number and has to wait to fill this gap. While, SCTP can deliver data to its upper layer protocol even if there is a gap in TSN if the Stream Sequence Numbers are in sequence for a particular stream. This event does not affect cwnd and only affect rwnd calculation.