Performance evaluation of different wireless network topologies using the IEEE 802.11 MAC Protocol
CHRIS BASIOS*, PANAGIOTIS KOSTARAKIS* and EVANGELOS PALLIS+
* Physics Department - Electronics, Telecommunications and Applications Laboratory
+Institute of Informatics and Telecommunications - Telecommunications Laboratory
*University of Ioannina
+NationalCenter for Scientific Research ‘Demokritos’
*Panepistimioupolis, 45110, Ioannina
+Ag. Paraskevi Attikis, 15310, Attiki
GREECE
Absract: WLANs are an emerging part of the current wireless communications providing high bandwidth to users in a limited geographical area. The IEEE 802.11 Standard is a widely used protocol for this kind of communication. In this paper, we estimated the performance of the above protocol for 3 different wireless networks; an ad-hoc network, a network that makes use of the LMDS technology and an infrastructure network. The performance evaluation was based on throughput values coming out of various simulations. Our main interest focused on the transfer of fixed data packets under a varying network load (depending on the number of users, the length of the packets etc.).
Key-Words: 802.11, DCF, throughput, ad-hoc, LMDS, infrastructure
1. Introduction
The main concept in our simulations was the throughput estimation for the case where every user that resides in the WLAN requests and downloads – using either TCP or UDP transport protocol - a fixed size message from a remote server. In the ad-hoc network both users and the server use the DCF access method – specified by the IEEE 802.11 Standard – in order to gain access to the wireless medium. In the LMDS model, both users and the AP (Access Point) of the wireless cell, which is connected via a point-to-point link to a wired server, use the DCF access method. Finally, in the infrastructure network the users use the DCF in order to request and download a message from the server that resides in a wired LAN. We have modeled various scenarios in order to investigate the network efficiency under different conditions. Thus, we have varied the offered load (by altering the number of the users, the arrival rate of the users’ requests to the server, the message length and the length of the MSDUs the message splits to) and we have also investigated the utilization of the fragmentation mode provided by IEEE 802.11 Standard.
The design of the simulation models and the analysis of their efficiency have been made by using the COMNET III simulation program. With this simulation tool we had the possibility to built our own network topology by using and altering many parameters of the main building blocks of the program - such as nodes, links, network devices etc. - that constructed our designed models.
2. The DCF access method and simulation basic parameters
In all simulations the access method that was used by the wireless users was the DCF (Distributed Coordination Function). DCF is the primary access method provided by the IEEE 802.11 Standard and is based on CSMA/CA algorithm. In the following lines we will give an overview of this algorithm.
Before a station initiates transmission of a data frame, it needs to sense the channel in order to determine whether another station is currently transmitting. The station can proceed with its transmission if the medium is determined to be idle for a time interval of DIFS (DCF Inter-Frame Space). After a data frame is successfully received at the destination, the receiver must send an acknowledgment frame (ACK), because the transmitter cannot determine whether a frame has been faithfully delivered to its destination by simply listening to the channel; the sender may not observe frame collisions the receiver detects with other senders not observable by the first (this is the so-called “hidden terminal problem”). To transmit the ACK, the receiver waits for the channel to be idle for another time interval of SIFS (Short Inter-Frame Space). If the transmitter does not receive an ACK packet within a certain period (ACK timeout), it presumes that the data frame is lost and schedules a re-transmission.
If the medium is busy upon transmitting a data frame or an ACK, the transmission must be deferred until the end of the ongoing transmission. In this case, a random backoff interval is selected, as follows. A backoff timer is set with a random backoff integer (BV) drawn from a uniform distribution over the interval [0,CW - 1], where CW (Collision Window) is an integer within the range of CWmin and CWmax. BV is the number of idle “slots” the station must wait until it is allowed to transmit. The value is decremented by one for each idle slot detected. The backoff timer suspends when the medium becomes busy before BV reaches zero. The timer resumes only after the medium has been idle longer than the designated inter-frame space interval. The station starts transmitting the frame when the backoff timer reaches zero. For each successive retransmissions, the value of CW increases exponentially (i.e. CWnew = CWold . 2 - 1), until it reaches and then stays at CWmax . CW will be reset to CWmin after a successful transmission. The backoff method is used to minimize collisions and maximize throughput at both low and high network utilizations.
All the wireless users access the wireless channel of each one of the 3 network topologies by using the basic DCF method provided by the 802.11 Standard, while at the physical layer they use the FHSS technique. The bit rate of the wireless medium is fixed to 2 Mbps.
The transport protocols we used are TCP and UDP. The congestion control algorithm implemented in TCP was selected to be the ‘Fast recovery’ algorithm. We also modeled the establishment and disconnection of a TCP connection using 2 fixed size (40 bytes) packets for every message (‘Open’, ‘Close’ packets), while the ACK packet acknowledging every data packet is 40 bytes long.
The main 802.11 parameters and their values used in the simulations are summarized in the next table.
Table 1: Parameters used for the simulation of the IEEE 802.11 MAC protocol
802.11 Parameters / ValuesBandwidht / 2 Mbps
Frame max / 2304 bytes
ACK timeout / 0.2 ms
Frame TxLife Time / 512 ms
Slot time / 0.05 ms
SIFS / 0.028 ms
FH dwell time / 128 ms
CWmin / 7
CWmax / 255
The main assumption that have been made for the wireless networks is that the ‘hidden-node’ problem is not addressed in the simulation models, while the wireless users are all in fixed locations (no mobility).
3. Simulation models and results
In this section, we present the 3 different simulation models and the individual scenarios being used. We, briefly, analyze the network topology and we demonstrate the throughput graphics coming out for each scenario.
3.1 Ad-hoc model
In this model we simulated the transfer of a fixed-size message from a wireless server to a variable number of wireless users with both users and the server residing in the same IBSS. Some simulations took place for 2 different environments; the first was an error-free environment and in the second we simulated the following basic delays and error factors:
- Packetize delay: 0.01 ms
- Buffer port delay: 0.05 ms
- Propagation delay: 1 μs
- BER: 10-3
- ACK BER: 0.112
Figure 1: Simulation model of the ‘ad-hoc’ network.
1st scenario
Every BSS STA (user) requests (1000 bytes packet) for a single time a 2 MB file from the server located in the same BSS. The requests take place in the first 120 sec of the overall simulation run time (3600 sec) according to a uniform distribution. Both users and server make use of TCP and UDP protocols (TCP have a fixed sized packet consisted of 1460 data bytes and 40 overhead bytes, while UDP packet is consisted of 1480 data bytes and 20 overhead bytes). The offered load in the network varies according to the number of users that join the ad-hoc network (the min. number of users is 1 and the maximum is 30).
Figure 2: Effect of number of users for different simulation environments.
2nd scenario
The requests’ arrival for every user in the BSS follow a Poisson distribution, making the interarrival time between the request messages follow an exponential distribution with a mean value of 600 sec. In this scenario - that was only modeled for error-free environment - we demonstrate the effect of the MSDU packet length in the user throughput. We used 3 different TCP-UDP packets by changing the data bytes of every packet (1000 bytes, 1460 bytes and 2000 bytes) that contract the whole 2MB message. Thus, the MSDUs coming out from the server will take 3 different values, with none of them exceeding the maximum allowable MSDU length (2304 bytes).
Figure 3: Effect of number of users for different MSDU lengths.
3rd scenario
In this scenario the number of users is fixed (equal to 10) and the varying parameter of the system is the length of the message being downloaded by all 10 users. Each request to the server is generated once (by every user) according to a uniform distribution. The request arrivals take place in the first 120 sec of the overall simulation run time (3600 sec). The message size varies from 1MB to 5MB, while the size of each transport protocol packet (TCP or UDP) remains the same equal to 1500 bytes (plus the overhead).
Figure 4: Effect of message size for different simulation environments.
3.2 LMDS model
The idea of the LMDS model was the investigation of the efficiency for a network topology where the users of a single wireless cell request and download a fixed size message from a service provider that resides in a remote wired LAN. The connection between the AP of the cell and the router that routes the packets up and down the server is implemented via an RF point-to-point link (2Mbps bandwidth). The service provider resides in a wired LAN, which uses the 802.3 10BASET technology, while the access method in the cell is carried out using the CSMA/CA algorithm for both uplink and downlink; that means that both the users and the AP make use of the 802.11 protocol. Some simulations took place for 2 different environments (error-free and error-modeling).
The parameters and their values used for the simulation of the IEEE 802.3 MAC protocol are according to the 802.3 Standard, while these of the 802.11 MAC protocol are the same with what was referred in the ad-hoc model.
In the error simulation environment we have modeled the same errors and delays with the previous model plus the BER of the wired medium - which was set to 10-5 - and the processing packet delay in both router and AP (Processing/packet: 0.0067 ms).
Figure 5: Simulation model of the LMDS network.
1st scenario
Every BSS STA (user) requests (using a 1000 bytes packet) and downloads a 2MB from the wired server. Every user that joins the wireless cell generates only one packet-request. The requests take place in the first 120 sec of the overall simulation run time (3600 sec) according to a uniform distribution. Both users and server make use of TCP and UDP protocols (TCP have a fixed sized packet consisted of 1460 data bytes and 40 overhead bytes, while UDP packet is consisted of 1480 data bytes and 20 overhead bytes).
Figure 6: Effect of number of users for different simulation environments.
2nd scenario
In this second scenario, we have modeled the 802.11 fragmentation mode. Firstly, we must mention that the arrivals of the users requests are modeled with exponential interarrival times with a mean value of 600 sec. The transport layer that was used in order to simulate the behavior of the fragmentation mode was UDP with a fixed size packet of 1500 bytes (1480 data bytes + 20 OH bytes). The fragmentation threshold was set in 712 bytes. In addition, we simulated the use of TCP in order to compare the efficiency of the 2 protocols when the load of the network increases (due to the increment of the wireless users).
Figure 7: The effect of fragmentation technique.
3rd scenario
In this scenario we investigate the effect of the message size in the users throughout under the 2 different simulation environments. The number of the users is fixed (equal to 10) and each request to the server is generated once (by every user) according to a uniform distribution in the first 120 sec of the overall simulation run time. The message size varies from 1 MB to 5 MB, while the size of each transport protocol packet (TCP or UDP) remains the same equal to 1500 bytes (plus the overhead).
Figure 8: Effect of message size for different simulation environments.
3.3 Infrastructure model
In this simulation model a 802.11 wireless BSS is interconnected via a wireless portal with a 802.3 wired LAN (10BASET technology). Both wireless and wired LAN consist of a varying number of users (1 up to 30). Every one of the above users will request and try to download a fixed data message from a server that is connected to the 802.3 LAN, using the TCP and UDP transport protocols. The main aim of this simulation model was to investigate the way the wireless users’ throughput is affected by the presence of the wired users. All the simulations took place for error-free environments.
Figure 9: Simulation model of the infrastructure network.
1st scenario
The message size that each user will try to download from the wired server is fixed and equal to 2 MB. Both wired and wireless users make a single request for the message and the number of both LANs’ users is incremented with the same rate (e.g. when in the WLAN reside 10 users, the same number of users shall reside in the wired LAN). The arrival of each request-packet will take place in the first 120 sec and will follow a uniform distribution.
Figure 10: Effect of number of users for different transport protocols.
2nd scenario
In this scenario we have modeled and simulated 2 different cases. In the first one we vary the number of wired users for a fixed number of wireless users, while on the second one we vary the number of wireless users for a fixed number of wired users. The arrival of requests from each station is modeled with exponential interarrival times with a mean value of 600 sec.
(a)
(b)
Figure 11: The effect of (a) increment of the wired users-the number of wireless users remains fixed and (b) increment of the wireless users-the number of wired users remains fixed.
3rd scenario
In this scenario we simulated the transfer of a fixed data packet from the wired server to 10 wireless users without the presence of any wired user. Both the users and the server make use of UDP with the number of the packet data bytes remaining fixed to 1480 bytes. The varying magnitude in this scenario is the message size, which varies from 0.01 MB to 5 MB. In order to estimate the throughput variation in association with the offered load, we used 2 different rates for the Poisson arrivals of every user request. Thus, the interrarival time is following an exponential distribution with 2 different values of 300 and 600 sec, respectively.
Figure 12: Effect of message size for different interarrival times.
4. Conclusions
The main conclusion coming out is the clearly better performance of UDP over TCP, whether the simulation environment used is error-free or not. This result has to do with the ARQ error mechanism provided by the 802.11 that makes the use of TCP congestion control and acknowledgement technique bouncy, especially for the networks that do not experience high load and collisions. As it was expected, TCP enters greater overhead than UDP because of the extra use of the ‘Open’, ‘Close’ and ACK packets. Another result for both ad-hoc and infrastructure networks is that the increase of the users turns to a throughput decrease for any kind of traffic; this means that downloading such large messages (1-5 MB) has a serious affection on the users’ throughput.
In particular, the use of larger fixed data packets (that construct the overall message) in the ad-hoc model is, generally, more efficient than the use of smaller ones. This is due to the ‘error-free’ simulation environment and the fact that when the 2 MB message splits in smaller MSDUs, more packets will have to contend for the wireless medium, thus increasing the probability of collisions. This explains and the decreasing of the throughput when the message size increases (from 1 to 5 MB).
In the LMDS model, the most remarkable conclusion is the limit estimation of the number of wireless users that makes the use of fragmentation either useful or not for the entire network. When the users of the wireless cell exceed the number of 15, the fragmentation mode leads to higher values of throughput. Despite the extra overhead bytes for every fragment, fragmentation increases the probability of successful transmission of the MSDU (because of the use of smaller frames) when the offered network load comes to higher values. The results of figure 2.3 show that for the error-free simulation model the offered load we applied in the network (10 users, one message request per user) leads in a slight throughput decrement despite the message size increment.
For the infrastructure simulation model, the existence of 10 fixed wired users does not affect in any way the throughput of the wireless users which remains fixed, independently of their number residing in the WLAN. The same wireless user throughout is slightly affected when the number of the wired users varies from 1 to 30. By varying the request generation rate and the length of the message size (figure 3.3), we turned out with 2 main goals. The first one has to do with the existence of 3 throughout levels; increment, saturation and decrement that lead to a clear limit of the best achievable throughput. The second one is the estimation of the maximum allowable message size value (1MB) over which the arrival rate affects the user throughput.