VOICE OVER WI-FI CAPACITY PLANNING

Transcription

1 VOICE OVER WI-FI CAPACITY PLANNING Version 1.0 <1> Copyright 2003

2 Table of Contents Introduction...3 Wi-Fi RF Technology Options...3 Spectrum Availability and Non-Overlapping Wi-Fi Channels...4 Limited channel availability with b and g...4 Expanded channel availability with a...6 Higher capacity networks require a higher number of channels...8 Capacity versus Range...8 Capacity Planning Example for a Mixed Voice/Data Network...9 Assumptions...9 Simulation...9 Access point placement...10 Sample handset deployment plot...12 Access Point Deployment Scenarios...13 Three access point deployment...13 Four access point deployment...14 The impact of increasing access point densities...15 Number of Simultaneous Calls per Access Point...18 The Impact of Data Traffic...21 Conclusion...24 Table of Figures Figure 1: Frequency reuse with b/g (2.4 GHz) and three (3) non-overlapping channels 5 Figure 2: 5 Spectrum availability difference between 2.4 GHz and 5 GHz...6 Figure 3: Improved frequency reuse possible with a (5 GHz)...7 Figure 4: Dense Deployments Require More Channels...8 Figure 5: Single access point coverage area and data rates...10 Figure 6: More access points added to increase coverage at desired data rate...11 Figure 7: Handset deployment plot...12 Figure 8: Three (3) access points support high-quality voice only 6% of the time...13 Figure 9: Four (4) access points deliver acceptable QoS approximately 30% of the time...14 Figure 10: Five (5) access points deliver acceptable QoS approximately 50% of the time...15 Figure 11: Six (6) access points deliver acceptable QoS approximately 70% of the time...16 Figure 12: Seven (7) access points deliver acceptable QoS approximately 80% of the time...17 Figure 13: Eight (8) access points deliver acceptable QoS over 99% of the time...18 Figure 14: Distribution of simultaneous calls per access point in voice & data network...19 Figure 15: Distribution of simultaneous calls per access point in voice-only network...20 Figure 16: Channel access with voice users only...21 Figure 17: Voice user channel access with background data traffic...22 Figure 18: Channel access improves with less voice users per access point...23 Figure 19: More access points (less users/ap) improves channel access...24 Version 1.0 <2> Copyright 2004

3 Introduction As voice-over-wi-fi emerges as a productivity-enhancing wireless application in the enterprise, the requirements placed on Wi-Fi infrastructure deployment planning increase from simple coverage and throughput considerations to detailed capacity planning. The issue of capacity planning is often overlooked with today s primarily data-oriented Wi-Fi networks, but, as in the cellular environment, capacity planning is critical when supporting a voice network with high Quality of Service (QoS). Most importantly, proper capacity planning should drive the decisions IT professionals must make in regards to Wi-Fi radio frequency (RF) technologies. Wi-Fi RF Technology Options Today, numerous manufacturers offer Wi-Fi certified products that conform to three (3) separate RF technologies. These three options are all derived from standards developed and ratified by the Institute of Electrical and Electronics Engineers (IEEE) Working Group for Wireless Local Area Networking standards. (Table 1) The Wi-Fi Alliance (WFA), a separate nonprofit industry association created to certify interoperability of products based on these IEEE standards, has subsequently established interoperability testing programs for each of these technologies. It is really the WFA testing and Wi-Fi certification process that ensures interoperability of wireless LAN products, not the standards themselves. Therefore, regardless of the technology implemented, enterprises must ensure that the products they ultimately deploy have passed interoperability certification through the WFA a b g Spectrum 5 GHz 2.4 GHz 2.4 GHz Maximum Channel 54 Mbps 11 Mbps 54 Mbps Rate Number of Channels Modulation (at the higher rates) OFDM CCK OFDM Table 1: Essential Parameters of Physical Layer Standards Version 1.0 <3> Copyright 2004

4 The majority of Wi-Fi deployments today are based on the b standard. Recently, products based on the newer g standard have been introduced, and have been gaining acceptance due to the combination of high over-the-air rate 1 (54 Mbps) and backward compatibility with existing b products. The a standard, which offers the same high data rate as g but does not offer backward compatibility or interoperability with b/g due to its operation in the 5 GHz frequency band, has experienced growth among enterprises that have recognized the need for high bandwidth and abundant, uncluttered spectrum. It is this issue of spectrum availability, which manifests itself within wireless LAN products in the form of non-overlapping channels that distinguishes a as the technology of choice for voice applications in high-density environments. Spectrum Availability and Non-Overlapping Wi-Fi Channels The number of non-overlapping channels is often overlooked in regards to wireless LANs, but is the primary determinant of overall system capacity. This is because access points configured on nonoverlapping channels can be aggregated to increase system capacity. For example, instead of all users within a given physical area having only one access point to associate with, if a second access point is installed in the same area but on a non-overlapping channel, those users will now be shared among the two access points, effectively doubling the system capacity. This is analogous to cellular system operators installing more base station radios to increase the number of subscribers they can support on their network in any given area. However, if access points are installed within radio range of each other and are not configured on nonoverlapping channels, then not only is there no aggregation benefit, but overall system performance is actually reduced. This is because, in an network, the packets from devices on overlapping channels will collide with each other. This will cause those devices to retransmit those packets using random time delays to avoid these collisions. This is known as having the devices back off to each other. This backing off results in time that is not spent sending data and, so, results in a decrease in system throughput. Even when the co-channel devices are not close enough to cause backing off, they will still elevate the noise floor in that channel. To overcome this type of co-channel interference, the radios must use lower data rates to reduce the packet error rate, and this, too, will result in a decrease in system throughput. Instead of doubling the capacity of a network, by adding a second access point on the same or adjacent channel, you end up with a lower-capacity network operating at reduced throughput. Limited channel availability with b and g While the pervasiveness of b and the high data rate and backward compatibility of g are valuable attributes and appropriate for many wireless LAN environments and applications, both technologies are insufficient for a high-density voice network. This is due to the spectrum availability limitations imposed by their operation in the 2.4 GHz frequency band. Radio frequency regulations governing the use of wireless LANs in the 2.4 GHz frequency band restrict systems operating within this band to three (3) non-overlapping channels 2. 1 Over-the-air rate refers to the maximum per-channel signaling rate of the radio. This should not be confused with throughput, which is the actual data rate available to an application and is dependent on radio efficiency and network protocol overhead. For example, although many in the industry will refer to b as delivering 11 Mbps of throughput, actual user throughput will typically be in the 5-7 Mbps range. 2 Wireless LAN products operating in the 2.4 GHz frequency band will indicate channels are available, depending on the regulatory domain. However, each of these channels overlap their adjacent channels, so only products configured for channels 1, 6, and 11 can be assured of not causing co-channel interference. Version 1.0 <4> Copyright 2004

5 This restriction imposes a significant system capacity limit on b/g deployments, as only three access points can be installed within an area of overlapping coverage. Looked at another way, the distance between access points on the same channel is relatively small, meaning that the same channels must be reused frequently throughout a given area, increasing the probability of co-channel interference in that area. (Figure 1) Distance between cells using the same frequency is very small Figure 1: Frequency reuse with b/g (2.4 GHz) and three (3) non-overlapping channels Version 1.0 <5> Copyright 2004

6 Expanded channel availability with a While a initially offered the advantage of up to 12 non-overlapping channels (depending on regulatory domain) along with a 54 Mbps data rate and clean spectrum, recent regulatory changes have opened up even more spectrum in the 5 GHz frequency band, creating the potential for over 20 non-overlapping channels. (Figure 2) Not all a products take advantage of all available channels. However, even products operating over a subset of the available a channels, for example those using eight (8) channels, have a significant capacity advantage over b/g products. Limited Limited spectrum spectrum and and three three (3) (3) channels channels available available in 2.4 GHz frequency band (802.11b/g) in 2.4 GHz frequency band (802.11b/g) 2.4 GHz 2.4 GHz Expanded spectrum and up to 23 channels Expanded spectrum and up to 23 channels available available in in 5 5 GHz GHz frequency frequency band band (802.11a) (802.11a) GHz GHz GHz GHz GHz GHz GHz GHz Figure 2: Spectrum availability difference between 2.4 GHz and 5 GHz Figure 3 depicts frequency reuse with a, and in this example only seven (7) channels are represented. The difference between reuse with only three channels, shown earlier, and this case is significant. With 8-20 non-overlapping channels available, a dramatically simplifies deployment in large-scale, high-density environments by facilitating much greater aggregation of access points and mitigating the impact of co-channel interference. Version 1.0 <6> Copyright 2004

7 Distance between cells using the same frequency is much larger, (only 7 channels represented in example) Figure 3: Improved frequency reuse possible with a (5 GHz) Version 1.0 <7> Copyright 2004

8 Higher capacity networks require a higher number of channels The conclusion one can draw from this discussion of spectrum and channel availability is that deploying high capacity networks demands a high density of access points. In Figure 4, the different colors represent different channels, and each colored section represents the coverage area of individual access points. You can see that the more colors (channels) that are available, the more you can increase the number of colored sections (access points) in an area without having to reuse the same color as frequently. The ability to deploy a high density of access points thus requires a high number of non-overlapping channels, which, for a given channel bandwidth, is made possible by access to more spectrum. Of the existing Wi-Fi technologies, a in the 5 GHz frequency band best supports these requirements for broad spectrum availability and a high number of nonoverlapping channels. Figure 4: Dense Deployments Require More Channels Capacity versus Range Wireless LAN access points are still often thought of more in relation to the coverage they provide, in other words, their range, than in the capacity they are able to deliver. Network capacity must be considered at the system level, and therefore the ability to aggregate access points becomes a much more critical factor than the range of an individual access point. In fact, in order to increase access point density in any given area, very often the range (output power) of each access point must actually be reduced in order to minimize co-channel interference as you add more access points into the area. Therefore, any differences between b/g and a concerning the range of coverage they provide, whether real or perceived, become irrelevant when considering a high-density deployment designed to support voice applications. The overriding factor in the voice applications environment becomes the increased channel capacity offered by a, as the range of each access point will intentionally be reduced to increase the access point density. Version 1.0 <8> Copyright 2004

9 Capacity Planning Example for a Mixed Voice/Data Network To illustrate the impact of capacity considerations in a high-density environment supporting both voice and data users, the following is an example of a capacity planning model for a highly idealized enterprise scenario. Though the parameters represent a simplified model, the lessons learned are applicable to an actual floor plan, density distribution, etc. In addition, some of the parameters, such as the user density and the Erlangs of traffic per user were chosen to clearly illustrate the impact of voice traffic on network performance. Actual values in enterprise voice deployments will vary. Assumptions The voice/data capacity planning example is based on the following assumptions: The environment being modeled is a single open office area within a large enterprise facility. The dimension of the area is 100 meters x 100 meters (10,000m 2 ). The model assumes there are 1,000 employees working within this area, for a population density of 0.20 employees per square meter. 50% of these employees have voice-over-wi-fi handsets. We are intentionally restricting voice traffic to use no more than 50% of the total channel capacity. 3 Average handset usage is 10 minutes per hour (0.17 erlangs of traffic per handset) The voice codec on the handset is 64 kbps with 20 millisecond voice frames The Wi-Fi infrastructure is a, deployed for coverage and subsequently adjusted for capacity according to the simulation In summary, the input to the simulation assumes 500 employees with voice handsets, each having a 17% chance of being active at any given time, so therefore there are, on average, 83 active handsets at any given point in time. Simulation The parameters assumed above are input into a simulation based on the following methodology: A Monte Carlo simulation 4 is run to disperse the active handsets throughout the area Each handset placed in the area is associated to the access point from it receives the highest signal to noise ratio. The carrier-to-interference (C/I) value 5 that each handset will experience is calculated. This C/I calculation is mapped to the data rate, frame error rate, and number of packet retries associated with the handset at its specific location. The amount of time each handset will occupy the wireless medium is calculated. 6 3 This allows us to simulate supporting high quality voice with a large amount of background traffic simultaneously. This does not represent a practical limitation, since in reality all available bandwidth will be used as needed, but it allows for a conservative method for modeling a mixture of high-quality voice traffic and background data traffic 4 A Monte Carlo simulation is a probabilistic model of a system involving an element of chance, through the use of random or pseudo-random variables. This makes it possible to understand the probabilities of different possible outcomes within that system. 5 Carrier refers to the desired signal, while interference refers to all undesired signals within the receiver s bandwidth, whether from thermal noise or an external RF source. 6 Calculation is based on the formula: 2*Retry_Factor*[(PHY_OVERHEAD+MPDU*8/Rate)+SIFS+(PHY_OVERHEAD+ACK*8/Rate)], where: the factor Version 1.0 <9> Copyright 2004

10 The contribution from all of the handsets is summed to determine the total time the wireless medium is occupied (total traffic load) The Monte Carlo simulation is repeated 1,000 times to simulate random dispersion of the handsets throughout the area The percentage of time each access point can handle the total traffic load is calculated. Access point placement A typical access point layout begins by determining a minimum data rate that will be required throughout the area to be covered. For this example, assume the access points are to be placed to provide a minimum of 36 Mbps in all locations. (The use of 36 Mbps is an example based on a reasonable compromise between bandwidth and access point quantity.) This represents a required minimum C/I of 23 db throughout the area. As shown in Figure 1, a single access point placed at the center of this area would have a coverage pattern as shown, and the average C/I would be approximately 21 db, below the required minimum C/I of 23 db. Therefore more access points must be placed in this area to increase coverage at the 36 Mbps data rate and thereby increase the C/I throughout the area. Figure 5: Single access point coverage area and data rates 2 accounts for the uplink and the downlink streams; Retry_Factor is a value >1 indicating how many additional times the packet may need to be retransmitted due to errors detected at the receiver; PHY_OVERHEAD is additional information sent with the data packet and added by the physical layer processes which reduces the user throughput below the total channel throughput; MPDU is medium access control (MAC) protocol data unit (MPDU), the unit of data exchanged between two peer MAC entities using the services of the physical layer (PHY); Rate is the data rate at which the MPDU is transmitted, in Mbps; SIFS is the short interframe space; ACK is the acknowledgement packet. Version 1.0 <10> Copyright 2004

11 If two additional access points are deployed, we find that the average C/I in the area increases to approximately 27 db. The coverage pattern of each of the three access points and their associated data rates are shown in Figure 6. We now find that the average C/I has increased to 27 db and we have increased the area within which we have achieved the required minimum data rate. Figure 6: More access points added to increase coverage at desired data rate While we have now increased the average data rate and C/I delivered by the network and therefore we are approaching an acceptable deployment from a coverage point of view, we have not yet taken capacity into consideration. This is the next step in the planning process, and requires modeling the number of users, their distribution throughout the coverage area, and the type of traffic they are generating. Version 1.0 <11> Copyright 2004

12 Sample handset deployment plot Capacity planning begins by modeling the distribution of users throughout the coverage area. For the purposes of this discussion, we are only taking the voice users into account. (Keep in mind that we are only permitting those voice users to occupy 50% of the channel capacity so that the resulting deployment will be able to support both high-quality voice and background data traffic simultaneously.) An example deployment of handsets according to the simulation above is shown in Figure 7. With the parameters discussed previously, there are 500 people with handsets, and those handsets have a 17% chance of being active at any given time, meaning that there will be approximately 83 active handsets, or an average of approximately 28 handsets per access point in a three (3) access point configuration. The next step is to calculate the voice channel usage as described and determine whether the system is, or is not, capable of supporting the voice traffic generated by the simulation. Figure 7: Handset deployment plot Version 1.0 <12> Copyright 2004

13 Access Point Deployment Scenarios The simulation above can be applied to a series of deployment scenarios based on various densities of access points. The results presented below illustrate a range of scenarios, from a minimum access point density deployment, representative of the limited 3-channel b/g case, to a higher density access point deployment simulating an 8-channel a installation. For each scenario, the percentage of time each access point is capable of supporting its voice traffic load is plotted. The shaded areas defined by the X-Y coordinates represent the coverage areas of the individual access points. The vertical axis, ranging in color from blue to red, represents the percentage of time each of the access points can support its total voice traffic load. The differences between the load levels of the shaded areas is due to subtle differences in the area (square footage) they are covering combined with the average number of handsets that the Monte Carlo simulation places within this area. Three access point deployment Figure 8 is a plot of the results from a deployment of three (3) access points each configured on a nonoverlapping channel. The scenario is again based on the assumptions of 50% voice users, each on active calls an average of 10 minutes per hour, as described earlier. In this case, each handset user is experiencing acceptable high-quality voice service in this area only 6% of the time. This means that 94% of the time, the users are either being denied access to the network in the case of a system with call admission control, or experiencing unacceptable voice quality in the case of a best effort QoS implementation. This scenario represents the highest capacity provided by an b/g deployment, due to the limitation of three (3) channels available in the 2.4 GHz frequency band. Figure 8: Three (3) access points support high-quality voice only 6% of the time Version 1.0 <13> Copyright 2004

14 Four access point deployment By adding a fourth access point configured on a fourth non-overlapping channel, the total capacity of the network is increased and therefore the capability of each access point to support its voice traffic is increased. In this case, the availability of two of the access points is increased to over 30%, but total system availability is still extremely low. (Figure 9) Figure 9: Four (4) access points deliver acceptable QoS approximately 30% of the time Version 1.0 <14> Copyright 2004

15 The impact of increasing access point densities Figure 10 through Figure 13 depict increasing access point densities, with a corresponding increase in the amount of time each access point is available to support voice services with acceptable QoS. It is not until eight (8) access points are deployed that this simulation shows an acceptable level of availability is achieved. Figure 10: Five (5) access points deliver acceptable QoS approximately 50% of the time Version 1.0 <15> Copyright 2004

16 Figure 11: Six (6) access points deliver acceptable QoS approximately 70% of the time Version 1.0 <16> Copyright 2004

17 Figure 12: Seven (7) access points deliver acceptable QoS approximately 80% of the time Version 1.0 <17> Copyright 2004

18 Figure 13: Eight (8) access points deliver acceptable QoS over 99% of the time Number of Simultaneous Calls per Access Point Note that as we deploy more access points, the average number of calls per access point declines respectively. In the eight (8) access point scenario we have an average of ten (10) simultaneous calls per access point. The reader must be careful not to interpret this as the maximum number of calls that each access point is capable of supporting. There are two important factors to consider. First, the network in this example has been designed for peak loading, meaning that in the eight access point scenario the average number of calls per access point is ten, but each access point may be required to handle over 20 calls, depending on the geographic distribution of handsets. (Figure 14. In the figure, the vertical axis in the graph represents the number of Monte Carlo events that correspond to each value on the horizontal axis.) Version 1.0 <18> Copyright 2004

19 Figure 14: Distribution of simultaneous calls per access point in voice & data network Additionally, recall that we have intentionally restricted the channel capacity for voice traffic to 50%. If we were to design the system without allowing background data traffic, keeping all other parameters the same, Figure 15 shows that with 8 access points each access point could support over 40 simultaneous calls, with an average distribution of over 20 calls per access point. Version 1.0 <19> Copyright 2004

20 Figure 15: Distribution of simultaneous calls per access point in voice-only network Deploying for capacity means that decisions must be made concerning whether to deploy for peak loading (the ability to handle the highest number of handsets that may congregate in any given area) which drives the need for access points higher, or simply deploying based on the maximum number of calls which can be supported per access point. This decision should be based on an assessment of the expected usage scenario in each area of the environment. Again, the example described in this paper assumed a 50/50 mix of voice and data throughout the facility. In an actual deployment, different areas of the enterprise may have different expected levels of voice usage vs. data usage, and the capacity plan can be adjusted accordingly. For simplicity, the example used in this paper is based on the conservative approach of designing for peak loading throughout the facility and allowing for equal levels of voice and data traffic. Version 1.0 <20> Copyright 2004

21 The Impact of Data Traffic For the purposes of simplicity, this example assumed a 50/50 mix of voice and data, in other words 50% of each access point s capacity was available for voice traffic. Wi-Fi equipment that has been designed to support toll-quality voice will include specific QoS mechanisms, such as pre e implementations, that prioritize voice traffic higher than data traffic. However, the presence of data on the network will impact the ability for voice packets to get access to the channel regardless of their priority level. As the amount of data traffic increases, the probability that a voice user can access the medium in the time required to maintain toll-quality voice is diminished proportionately. If it takes too long for a voice client to gain access to the channel, it will lead to unacceptable voice delay. As an example, Figure 16 shows a channel access in a best case voice-only scenario, assuming a channel fully loaded with voice clients and no data clients. For the purposes of illustration, we could assume the maximum delay that can be tolerated in a high quality voice network is 100 milliseconds. We can see that in the voice-only case we are well below this limit (90% of the packets take less than 10 milliseconds to access the channel). Figure 16: Channel access with voice users only To characterize the impact of data traffic on voice users in a mixed voice and data network, assume we now add in for example 50 Mbps of data traffic generated from five (5) additional data users. Figure 17 now shows how the average delay each voice user is experiencing in gaining access to the channel has increased. In this case, 10% of the voice packets are exceeding our 100 millisecond limit. This situation would obviously get increasingly worse as more data traffic is added to the network. Version 1.0 <21> Copyright 2004

22 Figure 17: Voice user channel access with background data traffic Obviously one way to mitigate the problem would be to limit the amount of data traffic allowed on the network. However, this is impractical in a typical enterprise environment. In fact, the opposite is more likely to be true, that is that the amount of data traffic on the network will be highly variable and unpredictable as data users perform various high bandwidth functions such as large file transfers. The alternative solution, and the basis of the scenario presented here, is to increase the number of access points (channels) available, and thereby reduce the probability of having an excessive number of active calls per access point in the presence of background data traffic For example, as we continue with the analysis of channel access delay in a mixed network, if we now reduce the number of voice users per access point by deploying more access points, assuming the same background data traffic, we can see in Figure 18 that the channel access delay for voice users has been reduced to an acceptable level (90% of the voice users are gaining access to the channel in less than 30 milliseconds). Version 1.0 <22> Copyright 2004

23 Figure 18: Channel access improves with less voice users per access point Figure 19 summarizes the impact of having more channels available in a mixed voice and data network. As more access points are deployed, the number of users per access point is decreased and therefore channel access is increased. Assuming that we require voice users to gain access to the channel in less than 100 milliseconds 90% of the time (circled region), the purple dot-dashed curve represents the most practical solution. In other words, limiting the network to voice users only (blue solid curve) is impractical, while trying to support the higher number of voice and data users (red dashed curve) results in unacceptable performance. Version 1.0 <23> Copyright 2004

24 Figure 19: More access points (less users/ap) improves channel access Conclusion High-quality voice-over-wi-fi requires increased capacity in addition to enterprise-wide coverage. The range of an access point becomes a relatively irrelevant factor in a high-capacity voice-oriented network, and system capacity becomes the critical issue in delivering voice services with high QoS to high densities of voice handset users. Increased capacity requires more non-overlapping channels b and g are limited to three (3) non-overlapping channels, while a offers up to 23 non-overlapping channels a is therefore the most appropriate RF technology available to support high-quality voice applications in high-density environments. March 2004 Version 1.0 <24> Copyright 2004