Performance Measurement over Mobile WiMAX/IEEE 802.16e Network∗ Dongmyoung Kim, Hua Cai, Minsoo Na, and Sunghyun Choi School of Electrical Engineering and INMC, Seoul National University, Korea {dmkim, chai, msna}@mwnl.snu.ac.kr,
[email protected]
Abstract WiBro(Wireless Broadband) is a Korean version of Mobile WiMAX/IEEE 802.16e system, which is designed for mobile broadband wireless access. Being a subset of IEEE 802.16e, WiBro employs orthogonal frequency division multiple access (OFDMA) and time division duplexing (TDD) schemes operating at 2.3 GHz bands. In mid 2006, the world first commercial Mobile WiMAX service, based on WiBro specification, started in Seoul, Korea. In this paper, we analyze the performance of commercial WiBro networks through traffic measurements. Many experiments are conducted in the various environments. We analyze the link capacity when the user datagram protocol (UDP) packets from multiple users fully utilize wireless links. We also analyze goodput performance with transmission control protocol (TCP) as well as round trip time performance. The measured performances are compared with those of HSDPA (High-Speed Downlink Packet Access) which is a competing system of Mobile WiMAX. We found that the RTTs of WiBro and HSDPA are very large compared with conventional data networks, e.g., Ethernet or Wireless LANs. Furthermore, it is shown that the user-perceived performance is limited by such long round trip times when TCP is utilized. It motivates us to improve round trip time performance of WiBro system. Finally, we analyze the VoIP performance and the performance when the user moves around the whole city.
1 Introduction Along with the advance of communication technology, the need for ubiquitous access to the Internet is increasing today. Many systems are being developed to support reliable high-speed data communication environment for mobile users. A prominent technology is Mo∗ This work was in part supported by the MKE (Ministry of Knowledge Economy), Korea, under the IT R&D program (2007-F-038-02, Fundamental Technologies for the Future Internet) supervised by IITA.
c 978-1-42 44-2100-8/08/$25.00 2008 IEEE
bile WiMAX/IEEE 802.16e [5]. WiMAX is defined for the worldwide interoperability of broadband wireless access by the WiMAX Forum, which promotes conformance and interoperability of the IEEE 802.16 standards. Mobile WiMAX is a mobile version of WiMAX, and it is based on IEEE 802.16e-2005 standard. In this system, some functionalities are newly adopted to support mobility. Mobile WiMAX has strengths in providing better mobility support compared with WLANs (Wireless Local Area Networks) and a higher transmission rate than conventional circuitbased cellular networks. WiBro (Wireless Broadband) service, which is based on Mobile WiMAX technology, is a portable Internet service in Korea. The funcrionalities of WiBro are defined by a Mobile WiMAX system profile, and it will be certified under one of the Mobile WiMAX system profiles. WiBro operates at 2.3 GHz bands with a 10 MHz channel bandwidth, and it employs orthogonal frequency division multiple access (OFDMA) and time division duplexing (TDD) schemes. Furthermore, WiBro provides a seamless data communication even when we move at a speed of 60 Km/h. In 2006, WiBro systems were deployed in Seoul, Korea, and the world first commercial service based on Mobile WiMAX/IEEE 802.16e commenced. Quite a few researchers have studied the performance of IEEE 802.16e systems intensively. However, most efforts are based on either computer simulations or numerical analysis, and both simulations and analysis have inherent limitations since they are developed from numerical models and the assumptions for realities. In this context, the performance evaluation based on measurement is very essential since we can obtain the realistic view on the system. Motivated by this, we present measurement results for public commercial WiBro networks in this paper. We present both system performance and single-user performance. We measured an aggregated goodput of multiple users in the saturated situation, i.e., the users always have packets to transmit. The aggregate goodput could be considered a system-wise performance, because we can estimate the overall link capacity of the system based on it.
Though the system level capacity is important, the high system capacity does not always imply an excellent communication environment for individual users. Delay performances such as round trip time (RTT) and delay jitter are also very important. Therefore, we also measured round trip times in various environments. Furthermore, many applications use TCP as a transport layer protocol. TCP has its own dynamic behavior to control congestion so that it is difficult to estimate the TCP performance in the real world. Accordingly, we analyzed TCP performance through experiments by adjusting TCP receive window sizes and file sizes. Based on the measurement results of TCP connections, we discuss how the delay of the network limits the single user goodput. The paper is organized as follows. We first introduce some related work in Section 2, and the measurement environments are described in Section 3. In Section 4, delay performance is analyzed, and the aggregated goodput in the saturated condition is presented in Section 5. Then, we discuss the performance of TCP connection over WiBro in Section 6. The performance with large scale mobility and the performance of VoIP application are analyzed in Section 7 and Section 8, respectively. Finally, the paper concludes with some concluding remarks in Section 9.
2 Related Work Nowadays, emerging wireless packet data systems, e.g., HSDPA and WiBro, are being commercially deployed in many countries. The performance of those systems has been investigated by many researchers. For example, the goodput and delay performance of WiBro and HSDPA are compared based on computer simulations in [9]. However, real performances in the live networks might be very different from analyzed ones because of many unexpected factors. Recently, there have been some efforts to evaluate the emerging wireless packet data networks based on measurements in live operational networks. In [6], the authors analyze the goodput and delay of HSDPA systems deployed in Finland. The performance metrics measured in HSDPA are compared with those of W-CDMA (Wideband Code Division Multiple Access) system, which is a precedent version of HSDPA. The delay and jitter characteristics of HSDPA are further analyzed in [8]. In [7], the authors concentrated on the TCP performance of UMTS (Universal Mobile Telecommunications System) live networks. The paper presents that the performance of TCP flow over the live UMTS network is strongly affected by system parameters such as RTT and TCP window size and so on. Compared with other systems mentioned above, the measurement report for the Mobile Wimax/WiBro system is very rare in the literature. To our best knowledge, our paper is one of the first, which address the live performance in the commercial Mobile WiMAX networks.
Figure 1. Measurement locations in SNU campus.
3 Measurement Environments Before presenting our measurement results, we briefly overview our measurement environments. We carried out our measurements in a commercial WiBro network, which is deployed in the campus of Seoul National University (SNU), Seoul, Korea. The service providers of WiBro and HSDPA used in our experiment are KT and KTF, respectively, which are sister companies. The corresponding product names are KT WiBro and KTF I-Plug, respectively. Fig. 1 shows the campus map of SNU along with the measurement locations, which are described below. One base station (BS) is deployed in the campus, and it is located about 200 meters away from our basic measurement spot, i.e., the building 132 in SNU. The experiments were carried out at the following five location/environment conditions. 1. Outdoor Center: The outdoor housetop of building 132. There exists a line of sight (LOS) from the base station. 2. Indoor Center: The 4th floor in building 132. Channel environment is poor though the building is closely located to the base station as there is no indoor repeater. 3. LOS Repeater: The 3rd floor in building 301. Channel environment is good though the building is located far from the base station because an indoor repeater is deployed at the vicinity of the measurement location, i.e., LOS environment from repeater. 4. Cell Edge: This spot is located about 1 kilometer away from the base station. The channel quality is not good even if it is an outdoor environment. 5. Mobile Condition: The channel environment the user experiences when moving along a circular road sur-
1
0.8
CDF
0.6
0.4
0.2
Figure 2. WiBro station to a wired host.
WiBro Outdoor WiBro Indoor HSDPA Outdoor HSDPA Indoor
0 80
rounding the campus an the average speed of 30 km/h. We assume that handoffs do not occur in this case.
120
140
160
180
200
Round Trip Time (ms)
(a) RTT performance comparison of WiBro and HSDPA. 1
0.8
0.6 CDF
For the sake of convenience, we use the above five terms for the rest of this paper in order to refer to the corresponding measurement location/environment. The first two environments above are the basic measurement environments so that all the measurements are conducted in the first two environments. On the contrary, only the RTT performance and the saturated UDP goodput performance are measured for the remaining three environments. In order to evaluate the performance of WiBro, we rely on three different tools: the first one is a commercial IP performance testing toolkit, called IxChariot [1], and the second second one is J-Perf [2]. IxChariot is used to measure response times, and J-Perf is used to generate packets. In addition, we utilized Wireshark [3] to capture and analyze the communicated packets. For WiBro host machines, we use five laptops and they are connected with the WiBro access network via Samsung SPHH1100 PCMCIAtype WiBro card.
100
0.4
Outdoor_Center Indoor_Center LOS_Repeater Cell_Edge Moving_Condition
0.2
0 80
100
120 140 160 Round Trip Time (ms)
180
200
(b) RTT of WiBro in the various conditions.
Figure 3. Round trip time performance of WiBro.
4 Delay Performance Delay characteristics of WiBro is noteworthy because it can significantly influence the overall performance of the users. The method we used to measure the round trip time is as follows. A small UDP packet is sent over the WiBro link from a laptop computer connected to the WiBro network to a desktop connected to the wired network. Then, the desktop sends back an answering message. Upon receiving the message, the laptop determines the round trip time. This procedure is repeated 25000 times, and then the result is analyzed. Fig. 3(a) shows the Cumulative Distribution Function (CDF) of RTT for both WiBro and HSDPA systems at the two basic locations. The magnitude of response time is little influenced by the channel statuses in the average sense, but some packets which experience long RTT are captured. The difference of RTT seems to appear because of the retransmissions at the wireless link. In this figure, WiBro and
HSDPA systems show a bit different characteristics. In the average sense, the RTT of HSDPA is shorter than that of WiBro. However, the distribution of RTT is more spread in HSDPA. The variation of RTT in HSDPA even increases inside building. In addition, the RTTs of WiBro system in three additional environments are also shown in Fig. 3(b) and Table 1. Three additional environments are LOS Repeater, Cell Edge, and Mobile Condition which are defined in Section 3. The RTT performance at LOS Repeater is similar to the performance at Outdoor Center. Furthermore, the RTT performance is not degraded a lot even at Cell Edge condition. The delay at the environment is shown to be very stable. On the other hand, in the case of Mobile Condition, the mean and variance of the RTT increase a lot compared with the other environments. Furthermore, some probe packets experienced very long RTT delay. From the result, we learn that the RTT performance is strongly correlated by the
Table 1. Representative values for round trip time (unit: msec)
Downlink,Outdoor Downlink,Indoor
12
Outdoor Center Indoor Center LOS Repeater Cell Edge Moving Condition HSDPA Outdoor HSDPA Indoor WiBro-to-WiBro WLANs
Minimum 49 44 133 135 50 98 98 185 8
Maximum 339 464 145 147 1316 570 2239 499 12
Goodput (Mbps)
10
Average 134 141 138 139 176 123 143 291 10
8
6
4
2
0 1
1.5
5 Multi User Performance with UDP saturation 5.1
Performance in a Saturated Condition
In this section, the link capacity of WiBro system is analyzed based on the UDP saturation experiment. The procedure to estimate the system capacity through the experiment is as follows. From one to five connections were made for both downlink and uplink. Even though the traffic generation rate of 20 Mbps by one transmitter is fast enough to saturate the wireless channel, we cannot conclude that the goodput achieved by a user is the total capacity of the system as the maximum resource for each user might be configured to be limited. Therefore, we increased the number of
2.5
3
3.5
4
4.5
5
4.5
5
Number of Users
(a) Aggregate goodput for downlink. 3.5
Uplink,Outdoor Uplink,Indoor
3 2.5 Goodput (Mbps)
channel fluctuation. It seems that poor but stable channel, e.g., Cell Edge, is better than fluctuating channel to provide good delay performance in the real system. In general, RTTs of WiBro and HSDPA are far longer than that of wireless LAN as shown in Table 1. To check whether this delay came from WiBro link or wired Internet link, we examined RTT added to each hop through the “traceroute” command. As a result, we find that the delay added on the wired Internet is as small as a few milliseconds, and most of delays are from the WiBro link. Because of this, RTT between two WiBro terminals are very large even though the two WiBro terminals are located in the same cell. Long delay caused by WiBro link might work as a great restriction for users. It influences the performance of realtime services such as VoIP (Voice over IP), and it can also cause the stall problem that restricts the use of bandwidth for each link when TCP is used. The VoIP performance on WiBro system will be discussed in Section 8, and the performance of TCP will be discussed in Section 6, respectively.
2
2 1.5 1 0.5 0 1
1.5
2
2.5
3
3.5
4
Number of Users
(b) Aggregate goodput for uplink.
Figure 4. UDP saturation performance with multiple users.
transmitters up to 5. When goodput is no longer increased along with the increase of the transmitter number, we conclude that the channel is saturated, and determine the aggregated goodput as the capacity of the system. We conducted all the UDP saturation experiments, which are presented in Section 5, at specific times of the day, i.e., from mid-night to early morning of week days, when the impact of high traffic load and interference from other users is likely to be minimal. According to the result of outdoor downlink environment in Fig. 4(a), we find that the WiBro system provides the maximum goodput of about 5 Mbps for one user. When three or more saturated users exist, the wireless channel seems to be fully utilized because aggregated goodput becomes almost the same beginning the three user case. Therefore, the system-level aggregated goodput attainable in the outdoor environment near the base station is estimated to be about 11 Mbps. In all four environments, the
Table 2. Saturated goodput performance DL (Mbps) UL (Mbps) Outdoor Center 10.558 3.093 Indoor Center 7.814 1.163 LOS Repeater 9.521 3.010 Cell Edge 4.534 0.834 Moving Condition 6.452 1.220
wireless link was not fully utilized when only one saturated user transmits packets. The KT WiBro system seems to restrict the use of the entire capacity by a single user even if there exist some remaining resources. Inside the building, the aggregated downlink goodput is reduced to about 7 Mbps, but it is still quite a good performance compared with HSDPA system. The saturated goodput for uplink communication is smaller that the donwlink case. Uplink capacity is shown to be about 3.1 Mbps at outdoor, and about 1.1 Mbps at indoor. In a poor channel environment, the goodput degradation of the uplink communication is much more severe than that of the downlink communication. For uplink, only 30% of goodput is obtained in the indoor environment compared with the one obtained in the outdoor environment. In this testing environment, the building is very closely located to the base station, but the uplink performance was not very good. Therefore, more research and development about repeaters and small base stations are required in order to provide good environments for uplink communication inside buildings. Next, the saturated goodputs in three additional environments are also shown in Table 2. In this environment we measured only the aggregated goodput when four users are saturated, because the link was always almost fully utilized in the other conditions. It is noteworthy that the performance in the vicinity of indoor repeater is comparable with the performance at the outdoor nearby base station. More than 90% of goodput is observed in the environment for both uplink and downlink communications. It means that the indoor repeater effectively compensates the loss in channel gains. It is also shown that the aggregated goodput in mobile environment is higher than the one achieved in the cell boundary. It is impressive that the user in mobile achieve better performance in terms of average goodput in spite of the poor RTT performance.
5.2
Coexistence of TCP and UDP Flows
In the meantime, an experiment was conducted to figure out how fairly resources were distributed between TCP and UDP flows when UDP flow aiming to monopolize the air resource. In the first scenario, a user who generates the TCP
Table 3. Coexistence of TCP and UDP flows (unit: Mbps)
UL DL
Single flow TCP 1.772 2.453
Two users TCP UDP 1.116 1.768 1.976 5.210
Single user TCP UDP 0.080 1.871 0.352 4.929
flow and a user who generates the saturated UDP flow were placed in the same network. In the second scenario, the TCP flow and the UDP flow were generated at the same time within one user. As shown in Table 3, it is commonly confirmed both in uplink and downlink that if the two users generate flows separately, the resources are fairly distributed to the TCP user too, despite a large amount of UDP traffic is generated by the other user. It again ensures that unlike FIFO (First-In First-Out) scheduling, basic activities to classify and manage packets by destination are conducted in downlink communication as well. On the contrary, if two flows are simultaneously generated within one user, the UDP flow occupies almost all resources and the TCP flow is hardly delivered for both uplink and downlink communications. It tells that fairness can be supported through fair distribution of resources among users, but that various flows are not distinguished within one user to distribute the resources. It is concluded that the current KT WiBro system make a connection for each user, not for each flow. Because of this system configuration, the QoS of real-time service may not be guaranteed when a user uses the other application simultaneously. We will discuss the problem again when observing the performance of VoIP application in Section 8.
6 Single User Performance with TCP 6.1
TCP performance with various sizes of receive windows
In this section, we test the performance that a single user achieves when TCP is utilized as a transport layer protocol. TCP is a representative transport layer protocol, and most of the time when the Internet is used, the TCP is used. TCP retransmission technique basically belongs to sliding window-based ARQ. In sliding window-based ARQ, all the packets within a window can be (re)transmitted regardless of the transmission result of other packets. Whenever the first packet in the window receives ACK, window slides one packet further. If the first packet did not receive ACK, the window freezes, and hence the packets which are waiting for entering the window can not be transmitted. This phenomenon raises the issue that the link capacity can not be
10
4.5
Downlink,Outdoor Downlink,Indoor Uplink,Outdoor Uplink,Indoor HSDPA,Outdoor HSDPA,Indoor
4
1
3 Goodput (Mbps)
Goodput (Mbps)
3.5
Downlink,Outdoor Downlink,Indoor Uplink,Outdoor Uplink,Indoor HSDPA,Outdoor HSDPA,indoor WCDMA(UL),Outdoor
2.5 2
0.1
1.5 0.01
1 0.5 0 0
10
20
30
40
50
60
0.001
70
0.1
TCP Receive Window Size (KB)
Figure 5. TCP goodput performance with various window sizes.
utilized even if the medium is idle. We refer to this as window stall problem. If the size of window is K, the maximal goodput that a user can get is limited to K/RT T due to this window stall problem. Furthermore, this problem can severely happen, especially, in the system which provides large bandwidth and long delay like WiBro system. As depicted earlier, TCP is basically sliding-windowbased ARQ, but it is special because the window size is adaptively updated based on the network status. TCP uses the smaller value between congestion window and receive window as the size of window, and it is possible to continue increasing congestion window if congestion does not occur, but the receive window is fixed at a certain value. Therefore, when the end-to-end bandwidth is enough, the size of the TCP receive window determines the actual window size. Windows XP recommends 17.52 KB as a basic size for TCP receive window over the Internet. TCP goodput was measured in various environment by changing the size of the TCP receive window ranging from 4 KB to 64 KB. The result is shown in Fig. 5. In the case of outdoor downlink transmission, the window stall problem limits the maximum goodput even when the window size is 64 KB which is the maximum window size of Windows XP. In the case of outdoor uplink transmissions, the window stall occurs when the receive window size is less than 24 KB. The performance is bounded by link bandwidth when the window size is lager than 24 KB. If a basic receive window size of Windows XP, which is 17.52KB, is used, the maximum downlink goodput for a single TCP flow is strictly limited to less than 30% of link capacity. Although link capacity increases through adopting the new technologies like MIMO, above-mentioned problems will still limit the maximum goodput of a single TCP flow.
1
10 100 File size (KBytes)
1000
10000
Figure 6. TCP goodput performance with various file sizes.
6.2
TCP performance with various sizes of files
In this section, the TCP goodput is measured when the various size of files are transmitted. The goodput is measured with changing the packet size from 12.5 Bytes to 395 KB. The TCP goodput was averaged after 100 measurements, but the one of HSDPA was averaged only 5 times. The results are shown in Fig. 6. Please note that the graph is drawn in dB scale. The TCP goodput linearly increases as the file size increases when transmitting a small files, and saturates when the file size exceeding some thresholds. For example, in the case of outdoor downlink communication, several hundred kilobytes of file size is required to efficiently utilize the wireless link. The result is produced due to the dynamic behavior of TCP. In TCP, it takes some amount of time to increase sending rate to the end-to-end bandwidth. Hence, file transmission can be completed before achieving the maximum bandwidth when the file size is small. Furthermore, because the congestion window is increased when TCP ACK arrives, it takes long time to increase the sending rate with long RTT. In WiBro system and HSDPA system, RTT limits the TCP performance in the wide range of file sizes, as shown in Fig. 6. We have shown the inefficiency of TCP flow caused by long RTT in the above two sections. It is obvious that reducing RTT is one of the most importance missions to improve user-perceived performance of WiBro networks.
4
TCP Gootput (subway) TCP Gootput (bus)
3.5
Table 4. Performance summary in the city Subway Bus Average RTT (msec) 162 164 Maximum RTT (msec) 1215 1109 Minimum RTT (msec) 45 45 Average goodput (Mbps) 1.739 1.912 Maximum goodput (Mbps) 3.557 3.341 Minimum goodput (Mbps) 0.245 0.302
Goodput (Mbps)
3 2.5 2 1.5 1 0.5 0 0
200
400
600
800
1000
1200
1400
1600
1800
Time (seconds)
Figure 7. User-perceived downlink goodput in public vehicles.
7 Performance Evaluation with Large Scale Mobility In this section, the performance of WiBro system when a user moves around the city is presented. Previously, testing is conducted at various circumstances, but they were controlled testing environments, and the tests are only conducted in a single cell. Therefore, it did not represent the actual live performance of a user in a daily life. In this study, the downlink goodput performance of WiBro system was evaluated by riding a subway (green line) and a bus (bus number 5412) in Seoul. These measurements were conducted once from 2 PM to 5 PM in a week day. Fig. 7 shows TCP goodput versus measurement time when the user moves around Seoul. In the actual environment, the goodput is determined from time-varying channel environment. The goodput is poor when the user moves into shadowed areas. We observe that the goodput inside subway changes periodically because the repeaters are installed at each subway station so that channel condition is poor between stations. Though the user suffers from a periodic shadowing, the connection was not disconnected during 30 minutes of measurement time, and more than 200 Kbps of goodput was provided in the worst case. Inside bus, goodput variation is less periodic than subway. Furthermore, more than 1 Mbps of goodput in the bus was provided for about 80% of the measurement time. RTT increased a lot in this mobile environment, too. Though the average RTT is increased only by about 20 msec, we observe much longer delay values in these scenarios. The distribution of RTT is spread so that the considerable amount (around 20%) of packets experienced RTT more than 250 msec. As a result, the representative values are summarized in Table 4. Though some performance degradation was ob-
Table 5. Skype VoIP statistics with a 95% confidence interval
Indoor Outdoor
Average Round-trip time (msec) 176.4 ± 9.4 167.2±6.3
served as stated above, the measurement results support the fact that the current WiBro system works well in general. The average goodput of 1.7 Mbps is a considerable rate, and the connection is never disconnected during the several tens of minutes. We conclude that the world first commercial Mobile WiMAX service is well provided in Seoul, Korea.
8 VoIP Performance over WiBro Among various types of multimedia applications, VoIP is perhaps the most prosperous application for the Mobile WiMAX network since the demands for voice communication continue to grasp the significant position in the market. However, many people still have concerns about how well WiBro networks can provide the voice communication. Therefore, we present the performance of Skype over WiBro. Skype [4] is one of the most popular VoIP applications today. We opened a Skype session between two WiBro users and observed the performance of VoIP session. We repeated talking via VoIP ten times, where each conversation lasted for more than one minute. From the perspective of the subjective conversation quality, the quality of VoIP was pretty good even in indoor since we could hear the voice clearly without any delay or interruption. By using built-in menu called “See the technical call information”, we can obtain VoIP statistics of Skype. At the end of each VoIP session, we recorded the RTT and summarized in Table 5. Though we are not sure whether the provided RTT is the exact value, it can be thought as an approximated value of real RTT. When the users use VoIP applications, it has been shown that most of the users are satisfied with the one-way delay which is less than 200 msec. The notified RTT from Skype is quite small so that WiBro users can use VoIP application
protected in the current configuration of WiBro system. In order to properly provide VoIP, which is an important application for the success of Mobile WiMAX/WiBro, some methods to distinguish VoIP flows should be implemented in the commercial networks.
9 Concluding Remarks
Figure 8. Experimental configuration for background TCP traffic.
Table 6. Performance of VoIP application with background traffic (unit: msec)
DL background traffic UL background traffic
3-user case 110 477
4-user case 80.2 80.6
smoothly in those environments. Then, we established background TCP connections, and measured the performance again. Fig. 8 shows the two methods to establish background connections. In both cases, there are three background TCP connections and one VoIP connection. The VoIP user maintains only VoIP connection in the first case, and we refer to the environment as the 4-user case. In the second case, the VoIP user maintains both VoIP connection and TCP connection, and we refer to the environment as the 3-user case. The background traffic is for both uplink and downlink. The 3-user case models the user who uses VoIP service while downloading or uploading file. Therefore, it is not a strange situation. Here, the experiment is done only in outside building. The results are shown in Table 6. For the 4-user case, the background traffic does not degrade the performance of VoIP session very much. However, in the 3-user case, the performance of VoIP session is severely degraded because of the uplink TCP background traffic. 477 msec of RTT provides very poor quality of VoIP service. The reason why VoIP session does not work well in the specific scenario is similar to the reason stated in Section 5.2. It seems that VoIP connection in the 4-user case can be protected because the resource is managed per user basis. We conclude that two connections within one user cannot be distinguished or
We analyzed the performance of WiBro, which is the first commercial Mobile WiMAX system, in various environments. We found that WiBro system provides the higher downlink goodput than existing competing systems such as HSDPA. Though the uplink goodput was much reduced at indoor environment, uplink performance is also quite good in general. However, we also found that long RTT of WiBro system causes some problems for the user-perceived performance. When TCP is utilized as a transport layer protocol, long RTT limits the maximum goodput for a flow. Because the restriction is severe when the packet size or TCP receive window size is small, actual link capacity can be utilized only with large sizes of TCP receive window and packets. In order to overcome this, the RTT performance improvement is highly desired. In spite of some identified problems, which need to be fixed in the future, the WiBro system performs a role for mobile broadband wireless access in Seoul, Korea. We found that seamless communication was being provided even when a user moved around the city within a vehicle.
References [1] [2] [3] [4] [5]
[6]
[7]
[8]
[9]
IxChariot. Available: http://www.ixiacom.com/. J-Perf. Available: http://dast.nlanr.net/projects/jperf/. Wireshark. Available: http://www.wireshark.org/. Skype. Available: http://www.skype.com/. IEEE std. IEEE 802.16e-2006, Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access SystemsAmendment for Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands, Feb. 2006. M. Jurvansuu, J. Prokkola, M. Hanski, and Pekka Per¨al¨a. HSDPA Performance in Live Networks. In Proc. IEEE International Conference on Communications (ICC), Glasgow, Scotland, June 2007. M. Kohlwes, J. Riihijarvi, and P. Mahonen. Measurements of TCP performance over UMTS networks in near-ideal conditions. In Proc. IEEE Vehicular Technology Conference (VTC), Stockholm, Sweden, May 2005. J. Rokkola, M. Hanski, M. Jurvansuu, and M. Immonen. Measuring WCDMA and HSDPA Delay Characteristics with QoSMeT. In Proc. IEEE International Conference on Communications (ICC), Glasgow, Scotland, June 2007. S. Shin, C.-K. Kang, J.-C. Kim, and S.-H. Oh. The Performance Comparison between WiBro and HSDPA. In Proc. Internationl Symposium on Wireless Communication Systems (ISWCS), Siena, Italy, Sept. 2005.
