Investigation of Timescales for Channel, Rate, and Power Control in a Metropolitan Wireless Mesh Testbed1

Transcription

1 Investigation of Timescales for Channel, Rate, and Power Control in a Metropolitan Wireless Mesh Testbed1 1. Introduction Vangelis Angelakis, Konstantinos Mathioudakis, Emmanouil Delakis, Apostolos Traganitis, Vasilios A. Siris. Institute of Computer Science, Foundation for Research & Technology-Hellas (FORTH), P.O. Box 1385, GR Heraklion, Crete, Greece. Abstract: We investigate the timescales for channel assignment, power control, and rate control in a metropolitan mesh network with links whose distances range from 1.6 to 5 Km. In particular, for channel assignment we consider a procedure that accounts for both intra-network and external interference, and investigate the timescale at which channels need to be re-assigned in order to achieve high performance. For rate control, we compare an auto-rate algorithm that adapts the transmission rate on a per-packet basis, with a fixed transmission rate scheme. For power control, we investigate the minimum transmission power required to achieve a high throughput, and how this power changes during the course of a day. Keywords: mesh network, testbed experiments, timescale, interference 1.1 Motivation and Methodology A cornerstone objective in the design of a multi-radio wireless mesh network is the efficient utilization of the limited wireless resource and the radio network infrastructure. Controlling key parameters of a wireless communications system, such as the channel assignment, the transmission rate, and the transmission power helps to achieve this objective. Interference is a key factor that can lead to reduced capacity and performance of wireless networks in general, and wireless multi-radio mesh networks in particular. Interference can reduce the achievable transmission rate of wireless interfaces, can increase the frame loss ratio, and can reduce the utilization of the wireless resource (spectrum), e.g., due to contention in distributed channel access protocols such as One method for reducing the interference is to appropriately select the channels of wireless interfaces. The channel assignment problem in wireless multi-radio mesh networks involves assigning a channel to each wireless interface. In addition to reducing the level of interference, channel assignment also affects the connectivity of wireless mesh networks since two interfaces within the transmission range of each other can communicate only if they operate on the same channel. Indeed, there is a trade-off between maximizing the connectivity and reducing the level of interference. This trade-off between connectivity and interference exists in the case where the connectivity between mesh nodes or wireless interfaces is not given prior to channel assignment. On the other hand, it is also possible that the connectivity graph is known a priori. In this case, the connectivity graph serves as a constraint that needs to be satisfied with the channel assignment. The connectivity graph between specific wireless This work was supported in part by the European Commission in the 7 th Framework Programme through project EU-MESH (Enhanced, Ubiquitous, and Dependable Broadband Access using MESH Networks), ICT , Corresponding author: V. A. Siris, tel.: , fax: , vsiris@ics.forth.gr

2 interfaces is typically known a priori in cases where directional antennas are deployed, such as the metropolitan wireless mesh network considered in this paper. The objective of rate control is to adapt the transmission rate to the channel characteristics in order to improve performance, in terms of throughput and packet transmission delay. The throughput depends on both the transmission rate and the packet loss ratio. The maximum throughput is not necessarily achieved by the maximum transmission rate, or by the transmission rate with the lowest packet loss ratio. For real time applications, such as voiceover-ip and media streaming, the performance depends primarily on the packet transmission delay. Hence, the target of a rate control algorithm depends on the specific application requirements. The power control problem in wireless networks is that of selecting transmission powers at each radio interface in the network. The problem is complex, but it is also important because the choice of the transmission power values affects many aspects of the operation of the network, since it determines: (i) the quality of the signal arriving at a receiver, (ii) the effective range of a transmission, and (iii) the amount of the interference it creates to third receivers. Because of these factors, power control is a cross layer issue affecting: the Physical layer due to (i), the Channel Access sub-layer due to (ii) & (iii), the Network layer since (ii) affects the network connectivity & topology, and finally the Transport layer as the effects of interference result in increased bit-error rate, which leads to packet drops. 1.2 Paper Contribution and Organization The contribution of this paper is to investigate the timescales for performing channel assignment, power control, and rate control in a metropolitan mesh network with links whose distances range from 1.6 to 5 Km. In particular, for channel assignment we consider a procedure that accounts for both intra-network and external interference, and investigate the minimum timescale at which channels need to be re-assigned in order to achieve high performance. For rate control, we compare an auto-rate algorithm that adapts the transmission rate on a per-packet basis, with a fixed transmission rate scheme. For power control, we investigate the minimum transmission power required to achieve throughput close to what is achieved with the maximum transmission power, and how this minimum transmission power changes during the course of a day. The key question we address is whether the aforementioned mechanisms should be performed in a small timescale, in the order of packet transmission time in the case of rate and power control and hours in the case of channel assignment, or on a much slower timescale, in the order of days or weeks. The rest of the paper is organized as follows: Section 2 describes the metropolitan wireless mesh testbed where the experiments reported in this paper were carried out. Section 3 describes the channel assignment procedure that is investigated and the corresponding results on the timescale for channel assignment. Section 4 presents the investigations on rate and power adaptation. Finally, Section 5 summarizes the presented work and outlines the conclusions drawn from it. 2. Metropolitan Wireless Mesh Network Testbed The metropolitan mesh network where the experiments were conducted covers an area of approximately 60 Km 2 and contains 14 nodes, Figure 1, among which six are core mesh nodes, [REF07]. The distance and antennas used for the links between core mesh nodes are shown in Table 1. Wireless interfaces implement the IEEE a standard, and are assigned static IP addresses. The mesh network is connected to a fixed network through two nodes (K1 and K4). Each multi-radio mesh node consists of a mini-itx board (EPIA SP 13000, 1.3 GHz C3 CPU, 512 MB DDR400 memory). A four slot mini-pci to PCI adapter (MikroTik Router-BOARD 14) holds four a/g mini-pci adapters (NMP-8602 Atheros-based High Power dual band a/b/g). The mini-itx runs Gentoo 2006 i686 Linux ( kernel) with the MadWiFi driver (version 0.9.2).

3 An important feature of the testbed the separate management and monitoring network, which is possible by including an independent a client in each mesh node. This unique feature of the testbed allows remote assignment of different channels to the core links in the network, without losing connectivity or requiring tight synchronization between the corresponding nodes. Additionally, to enable remote recovery of the mesh node s mini-itx board in case it crashes, each node contains an intelligent remote power switch (Dataprobe iboot); the remote power switch supports off/on power switching through a web interface, and timed power reboots based on the results from the power switch pinging other devices (the mini-itx board or some remote device). Figure 1: The experimental metropolitan wireless multi-radio mesh network. The seven a links considered in the experiments involve the five core nodes K1-K5. Nodes M1-M8 are used solely for management & monitoring. 3. Channel Assignment Table 1. Links between the core mesh nodes Link Distance (Km) Antennas (all are panel) K1 K dbi - 21 dbi K1 K dbi - 21 dbi K2 K dbi - 19 dbi K4 K dbi - 21 dbi K4 K dbi - 19 dbi K2 K dbi 26 dbi K4 K dbi 26 dbi 3.1 Channel assignment procedure In this section we discuss the channel assignment procedure introduced in [REF08], which has three basic components, Figure 2: (i) the interference model, (ii) the link ordering, and (iii) the channel selection metric. An important requirement for channel assignment is to consider both the interference between links belonging to the mesh network (intra-network interference), and from sources external to the network (external interference). External interference can originate from both and non sources. Figure 2: The three components of the channel assignment procedure.

4 One approach for capturing intra-network interference is the Multi-Point Link Conflict Graph (MPLCG) presented in [REF08]. A vertex in the MPLCG represents a multi-point communication link, which is a set of interfaces that communicate with each other; all interfaces belonging to the same multi-point link should be assigned the same channel. An edge between two vertices in the MPLCG indicates that the two corresponding links interfere, hence cannot be assigned the same or neighbouring channels; the latter is enforced because there can exist interference between adjacent channels, even in the case of IEEE a. In [REF08] the MPLCG approach is compared with an approach where test-traffic is generated on links that have been assigned channels; once test-traffic is generated on a link that has an assigned channel, all links that are subsequently considered for channel assignment are able to measure the actual interference from that link. The additional contribution of this paper compared to [REF08] the investigation of the minimum timescales over which channels need to be re-assigned, in order to achieve high performance. A second important issue for channel assignment is the order in which links are considered for channel assignment. The channel assignment problem in mesh networks with multi-radio nodes is known to be NP-hard, e.g. see [SGD07]. For this reason we consider a heuristic approach where channels are assigned to links according to some predefined order, similar to [RC05], [RBAB06]. Such an approach can be followed in a centralized channel assignment scheme, which is realistic for moderate-sized mesh networks, when they are managed by a single authority. One alternative is to order links based on their distance from a fixed network gateway [RC05], [RBAB06]; this is based on the assumption that links closer to the gateway are more important since they forward traffic to and from the wired network. The final step of the procedure involves the channel selection, which for each link selects the channel to be assigned to it in a greedy fashion, based on the best value of the chosen metric. In [REF08] we investigate the following three metrics: 1) one-way SNR, 2) two-way SNR, which is the average SNR on the two interfaces belonging to the same link, and 3) round-trip delay. The above channel selection metrics can be measured online, and can capture the level of interference from external sources, both and non ; other approaches to channel assignment capture only interference between internal links [SGD07], [KMPR07], or external interference solely from sources [RBAB06]. The two SNR metrics capture adjacent channel interference, but do not capture MAC-layer contention between interfaces assigned to the same channel. On the other hand, the round-trip delay metric can capture interference due to both adjacent and cochannel interference, since it is influenced by MAC layer contention. Table 2: Channel Assignment Algorithm. The pseudo-code for the channel assignment procedure is shown in Table 2. Line 6 in the algorithm considers, for each vertex v V, only the channels that have a one channel separation from the channels that have already been assigned to another vertex u V, for which there is an edge with v in the multi-point link conflict graph. Among these channels, the one with the best metric is selected in line 7. The channel assignment algorithm is implemented in a stand-alone module that collects measurements from the links of the metropolitan wireless mesh network, using the separate man-

5 agement and monitoring network described in the previous section. The same management network is used to instruct the assignment of channels to the wireless interfaces. 3.2 Results for Channel Assignment Figure 3 compares the average packet delay achieved with the proposed channel assignment procedure using the Multi-Point Link Conflict Graph, with an interference-unaware approach and with a lower bound of the average packet delay; the lower bound was computed considering the optimal channel selection independently for the following link pairs: K4-K2 and K4-K5, K1-K2 and K2-K5, K1-K2 and K1-K3, while considering the link K4-K3 alone. For each pair of links, the optimal channels where selected by using a brute force search of the best channel pairs, among the 11^2 possible pairs since there are 11 channels for outdoor use in a, while the other interfaces were down. The results show that the proposed channel assignment procedure achieves an average packet delay that is 9% higher than the lower bound, while the interferenceunaware channel assignment procedure achieves an average packet delay that is 33% worst than the delay achieved by the proposed procedure; this indicates that there can be significant performance gains when the interference is taken into account for channel assignment. Figure 4 compares, for an interval of two weeks, the average packet delay when the channels are selected once at the beginning of the two week period, with an adaptive approach where new channels are selected once every day. The proposed channel assignment procedure was used in both cases. The results show that the fixed approach achieves an average packet delay that is within 9% of the adaptive approach. This suggests that, under normal conditions, there are no significant gains in performing channel assignment in a timescale smaller than 1-2 weeks. Figure 3: Comparison of proposed channel assignment with multi-point link conflict graph with a lower bound of average packet delay, and an interference-unaware channel assignment procedure. Figure 4: Comparison of fixed channel assignment for two weeks, with an adaptive approach which selects new channels every day.

6 4. Power & Rate control 4.1 Rate control in metropolitan wireless links over long time-scales In this section we present results from our investigation of rate control in the metropolitan wireless multi-radio mesh network testbed. Our objective is to compare the throughput that is achieved using a fixed transmission rate scheme, with the throughput achieved with MadWifi s SampleRate algorithm, which we will refer to as auto-rate scheme, that adjusts the transmission rate on a per packet basis. The traffic was generated using the Iperf tool. The graphs in this section report values averaged over intervals of two minutes. Figure 5 shows the throughput achieved on a specific link, with the auto-rate scheme (horizontal straight lines) and fixed transmission rate for different transmission powers. Observe that a higher throughput is achieved by using a fixed rate scheme, if the fixed transmission rate is appropriately selected. Moreover, the figure shows that a higher improvement is achieved in the case of lower SNR values: when the transmission power is 15 dbm the improvement is 25%, whereas when the transmission power is 13 dbm the improvement rises to 70%. Figure 5: Throughput for auto-rate and fixed transmission rate. Link K3-K4, distance 3.3 Km. Figure 6 compares the throughput that is achieved with auto-rate and with a fixed transmission rate, for a link with a smaller distance (2 Km) and better quality, compared to the link in Figure 5. Here, for high quality (which corresponds to high SNR values), the maximum throughput achieved by fixed rate is similar to the throughput achieved by the auto-rate scheme. However, for links with low SNR values (transmission power 5 dbm), the improvement can be up to 200%. Figure 6: Throughput for auto-rate and fixed transmission rate. Link K2-K3, distance 2 Km. Figure 7 shows the throughput for auto-rate and fixed rate, at different hours of the day. Observe that improvements achieved by fixed rate remain relatively the same during the course of a day.

7 Figure 7: Throughput for auto-rate and fixed transmission rate. Link K1-K3, distance 4.9 Km. The conclusions drawn from the experimental results in this section are the following: For high quality (high SNR) links, fixing the rate to the highest transmission rate achieves similar throughput as SampleRate. For low quality (low SNR) links, an appropriately selected fixed transmission rate can achieve significantly higher throughput compared to SampleRate. Finally, a fixed transmission rate scheme shows similar performance over long timescales (day). Hence, for metropolitan mesh links, the transmission rate does not necessarily need to be adapted over smaller timescales. The above suggests that the adaptation of the transmission rate in small timescales (e.g., of the order of packet arrivals), which is typically followed by all auto-rate algorithms, is not necessary, and can even reduce performance, for long distance metropolitan wireless links. 4.2 Joint Power and Rate control Next we investigate power control in metropolitan wireless links. The advantage of reducing the transmission power is the reduction of the interference that is created. The objective of the experiments in this section is to identify the minimum transmission power that achieves close to maximum performance, and investigate how this minimum transmission power varies with time. Figure 8 reports the packet loss probability as a function of the transmission power, for both auto-rate and fixed transmission rates. The key observation from this figure is that the same low packet loss behaviour is achieved with transmission power 10 dbm, which is significantly lower than the maximum transmission power of 15 dbm. A similar conclusion can be drawn from Figure 9, which shows the throughput for different transmission powers. Figure 10 shows, for different times-of-day, the transmission power required to achieve throughput above the 94% of the maximum throughput. This figure shows that the necessary minimum transmission power does not change significantly throughout the day. The corresponding throughput that is achieved for these values of the transmission power is shown in Figure 11. The conclusions drawn from the power control experiments are the following: The transmission power can be significantly reduced, without a large impact on the achieved throughput. The minimum transmission power to achieve some minimum performance does not significantly change throughout the course of a day, hence adjustment of the transmission power can occur on long timescales.

8 Figure 8: Packet loss probability for different transmission powers. Link K3-K4, distance 2 Km. Figure 9: Packet loss probability for different transmission powers. Link K3-K4, distance 2 Km. Figure 10: Transmission power for achieving Figure 11: Throughput achieved by the throughput above 94% of the maximum throughput. transmission power shown in Figure Conclusions The goal of this paper was to investigate the timescales for performing channel assignment, rate and power control in a metropolitan wireless mesh network with links whose distances range from 1.6 to 5 Km. For channel assignment, our results suggest that there are no significant performance gains when the assignment is performed in a timescale shorter than weeks. For rate control, our results show that in metropolitan wireless links, the adaptation of the transmission rate on very small timescales, on a per packet basis as is commonly the case with widely used auto-rate algorithms, does not necessarily achieve higher performance compared to a fixed transmission rate scheme, when the transmission rate for the latter is appropriately selected. Moreover, for low-quality links, e.g., links with a small signal-tonoise ratio, auto-rate algorithms can reduce the throughput significantly. Finally, the experimental results on power control show that the transmission power can be reduced considerably, before a significant impact on the achieved throughput is observed, and the minimum transmission power for achieving a target performance in terms of throughput does not change significantly in timescales on the order of hours. References [REF07] [REF08] [KMPR07] Paper details removed for anonymity. Paper details removed for anonymity. B. Ko, V. Misra, J. Padhye, and D. Rubenstein, Distributed Channel Assignment in Multi-Radio Mesh Networks, in Proc. of IEEE WCNC, 2007.

9 [SGD07] [RBAB06] [RC05] [RW07] A. P. Subramanian, H. Gupta, and S. R. Das, Minimum-Interference Channel Assignment in Multi-Radio Wireless Mesh Networks, in Proc. of IEEE SECON, K. N. Ramachandran, E. M. Belding, K. C. Almeroth, and M. M. Buddhikot, Interference-Aware Channel Assignment in Wulti-Radio Wireless Mesh Networks, in Proc. of IEEE INFOCOM, A. Raniwala and T. Chiuch, Architecture and Algorithms for an IEEE Based Multi- Channel Wireless Mesh Network, in Proc. of IEEE INFOCOM, A. H. M. Rad and W. S. Wong, Joint Channel Allocation, Interface Assigmnent and MAC Design for Multi-Channel Wireless Mesh Networks, in Proc. of IEEE INFOCOM, 2007.