VeriWave Inc. 8770 SW Nimbus Ave. Beaverton, OR 97008 http://www.veriwave.com Tel: (503) 473-8340 Fax: (503 473-8351 Version 2 May 2011
TABLE OF CONTENTS
INTRODUCTION .... 3 PREVIOUS PRACTICES AND PROBLEMS..... 3 MISGUIDED FOCUS ON RF SIGNAL STRENGTH ................................ 4
OPTIMIZING FOR COVERAGE AND NOT CAPACITY .... 5 TESTING IGNORES HIGH VALUE CLIENTS . 5 RESULTS DEPEND HEAVILY ON TECHNICIAN PERSONALITY ..6
NEW SITE DEPLOYMENT STRATEGY..... 9 PREPARING TO GO ONSITE...... 11 LOCATING THE WAVEDEPLOY ETHERNET SERVER OR EF1101 DEVICE .... 12
DEFINING LOADS AND SERVICE LEVEL AGREEMENTS .....14
TESTING ADJACENT SECTIONS WITH OVERLAPPING CHANNELS . 22 COMPLETING THE BUILD OUT.. 22 NON-TECHNICAL BEST PRACTICE CONSIDERATIONS..................... 23
IT IS NEVER AS GOOD AS YOU THINK .... 23 TEST IN THE LAB FIRST BEFORE A LARGE DEPLOYMENT ............. 23 APPLY PROCESS.... 23 PREPARE BEFORE GOING ONSITE ..... 24 REPORT THE RESULTS..... 24
CONCLUSION..... 24 WORKS CITED... 25 APPENDIX A: COMMON CORRECTIVE ACTIONS RESLTING FROM ADOPTING THESE BEST PRACTICES ... 26 APPENDIX B: ABOUT VERIWAVE...................... 28
WHAT SETS VERIWAVE APART?........... 28 ONLY VERIWAVE..............................28
INTRODUCTION
After years of serving as a nice-to-have hospitality solution, 802.11 is being thrust into the forefront of all major service provider networks as a solution for cellular data offload. Customers accept and even embrace this offload solution when it provides them with an improved mobile experience. Implemented properly, this solution improves the customer experience by increasing the mobile networks capacity and performance while simultaneously lowering the cost to deliver data for the service provider. The risk to service providers of using WiFi for cellular offload is that unsatisfactory user experiences with WiFi now result in a loss of high margin smart phone customers. A single smart phone subscriber accounts for thousands of dollars in revenue and hundreds in profit for a single smart phone contract, and potentially represents much more in a quad play bundled solution. Service pr oviders must make every reasonable effort to acquire and retain these premium customers in order to sustain and grow their business. Users have become sophisticated and now actively seek out the service provider that they believe offers the best overall network for their smart phone. They are quick to switch if they are unhappy with their existing service providers solution. The presen t reality is that subscribers are no longer are willing to accept poor WiFi performance with a simple shrug. A poor WiFi experience is a poor smart phone experience, and customers expect their smart phone to achieve WiFi performance that is comparable or improved to when it is operating on a 3G or 4G network. This document examines the best practices that the leading service providers are adopting to upgrade their WiFi deployments from a hot spot convenience service to a critical 3G / 4G offload infrastructure solution. It examines previous deployment practices and highlights the critical limitations that are forcing the leading service providers to adopt a new solution for deploying WiFi as a cellular offload solution. Finally, it describes how these leading service providers are upgrading their WiFi deployment practices to address these concerns and delivering a high quality solution that satisfies the needs of cellular offload over WiFi.
Conducting a site assessment in realistic network conditions may be impractical or impossible. For example, when a WLAN is deployed and users have not yet moved into the facility, it is impossible to create the entire planned traffic load one would expect. By utilizing WaveDeploy Expert and the WF1101 Traffic Generator, the network can be loaded with traffic from hundreds of golden clients that emulate the traffic loads users will place on the network when they start using it. Not only does this expedite testing, it allows the tester to detect any deficiencies in the network or in mobile-client interaction before actual users move in and experience difficulties.
It is highly recommended that any proposed SLAs be tested in the lab environment using the WaveTest system from VeriWave. The WaveTest solution is a powerful chassis-based system that allows users to create thousands of users in a lab environment before a deployment is conducted. Using this equipment enables carriers to perform the following tasks prior to deployment: Perform bakeoff testing between competing equipment providers prior to purchase Determine the application-level SLAs that the AP is capable of delivering prior to deployment Scale a solution to thousands of users to determine if there are any breakdowns in system performance (APs, controllers, servers, etc.) as the solution is scaled Test alternative deployment configurations for feasibility After equipment has been deployed into a live production network, the WaveTest is useful for: Recreating field-detected issues in the lab so that they can be addressed and verified fixed Regression testing new releases of infrastructure equipment prior to deployment in the live network Interoperability testing new client devices or upgrades to existing devices to identify any potential issues Determine if new overlay services can realistically be supported by the network prior to live deployment The combination of WaveTest and WaveDeploy is powerful to service providers because the solutions share a common set of measurement technologies that allow measurements made in the lab and field to be very strongly correlated. For large, high value deployments, this combination of solutions enables the engineering team to work hand-in-hand with the field team to ensure the best possible solution. A comprehensive discussion of the WaveTest solution is beyond the scope of this document. The reader is encouraged to go to www.veriwave.com or to contact VeriWave at +1-503-473-8350 for more information.
With everyone knowing what to expect and with the design well-vetted, it is generally fine to go forward with the remainder of the build out. Each new section should be tested for capacity, SLA compliance, and coverage, but this testing is now very quick to complete. Note that at each step along the way, it is possible to obtain a report as well as all of the detailed data associated with the testing. These reports are critical to knowing that the installation work was done correctly and completely. At the risk of belaboring the point, a test should be run after each incremental change during the design and a report generated or saved to document the results. If issues later come up after the deployment, these reports serve as a powerful baseline for determining what has changed. Upon completion of this process, you will know the level of performance that can be offered to your customers and delivered with confidence. It is easy enough to monitor the trends at the sites over time to determine when the usage patterns exceed the design points and know that it is time to upgrade the network before your customers start to complain too. This process allows service providers to leverage the high performance and low cost of WiFi to deliver an outstanding customer experience to their highest margin mobile customers.
10
PREPARING TO GO ONSITE
There is no better message when it comes to these engagements than the Boy Scout motto of Be Prepared. There are a number of details that can be easily addressed prior to going onsite, and there are a number of potential directions that a deployment test can take if the results are unexpected. It is best to eliminate as many potential problems before going onsite and plan for things to go wrong when you get there. It can be extremely frustrating to arrive onsite and realize that the critical personnel are unavailable to configure the network, that you do not know the security settings, or that you forgot to charge the battery before you left since you assumed that wall power would be readily available. Planning for these items ahead of time, and communicating the necessary support to others in the project, is critical to minimizing the cost and time involved with the deployment testing. Start by obtaining the necessary background information of the network and facility. Make sure you obtain image files for the site that will be tested and be clear with everyone regarding where the first section to be tested is located. Make sure you have access to get onsite so that the security guard does not turn your team away when you show up. Then make sure you have the information about the network itself. Make sure the team knows the SSID of the network that will be tested, the security settings, and the security credentials such as certificates or passwords that may be required. Note the intended install location of the APs, their associated BSSIDs and channels on one of the maps. Distribute this map to everyone on the team and especially the installers! It is often necessary for service providers to deploy in an environment where customers are active so some consideration should be given to restrict the network access to only those involved in testing. Hidden SSIDs or security are often used to restrict access. MAC address filtering can also be used, but is generally not advised when testing at scale using the WF1101 appliance due to the large number of MAC addresses involved. Some client devices do not function well with hidden SSIDs. Some client and network devices perform worse with security enabled than when they are open with no encryption. The best approach to address this issue is verify proper operation ahead of time in the lab using the same equipment as will be used in the assessment. Install all of the necessary WaveDeploy and WaveAgent resources on the target devices before going onsite. Make sure you have the latest version of the application by downloading it from http://www.wavedeploy.com. The WaveAgents for all of the platforms except Apple iDevices are included in the installer and can be found in the Start menu. The iDevice versions of WaveAgent must be downloaded from the Apple App Store to the iDevice using iTunes. Make sure you activate the WaveDeploy application licenses, including the support license if purchased. Make sure that you can run a Pro Assessment by following the example in the help file and verify that you know how to include all components. Clarify the difference between a speed test and concurrent traffic. Note that it also helps to become familiar with the smart phone behaviors. Many of these devices will sleep very quickly to conserve power, roam aggressively to maximize performance, or conversely roam very little to minimize risk of roaming failures and service interruption, and have limited data handling capabilities. Understanding the behavior of these devices before going onsite enables the tester to much more quickly pinpoint abnormal operation and find workarounds for unwanted device behaviors. If you will be using the WaveDeploy Expert solution then make sure you have updated the firmware and know how to connect the device to the network under test and run an assessment. Make sure that you are familiar with how to configure the WF1101 before going onsite. Make sure you know how to run an ecosystem test, view results, and download and view a capture file from both the WF1101 and the EF1101 appliances. The final equipment preparation involves making sure you have all of the necessary cabling, antennas, chargers, batteries, etc. to conduct the test. Charge the batteries before you go as power is not always readily available.
11
Figure 3
Figure 3 shows a typical controller-based hot spot deployment. There are two specific functions that commonly play into the choice of what equipment, topology, and configuration to use: Web authentication Firewall settings Web authentication refers to the mechanism whereby a user must open his browser, and is presented with a page where billing information can be provided and terms and conditions are presented that must be accepted prior to using the network. There is a specific challenge with web authentication because there is no supporting standard. Each vendor implements web authentication differently, and the implementations can vary between products from the same manufacturer or even between releases in the same product family. It is easy for people to interpret these screens and decide what information to provide and where they should click. It is a much more difficult task to automate this process. If possible, it is best to choose a configuration that circumvents the web authentication mechanism. If web authentication must be used then one can always use WaveDeploy Pro which allows up to 10 client devices in the test. The tester would navigate the web authentication procedure prior to initiating testing. Firewalls can also be a challenge to testing. WaveDeploy utilizes a distributed protocol that communicates using UDP on port 18100 by default. If a firewall is in use then a rule must be created to allow this traffic to flow.
12
In practice, we have found that the most ideal situation is when the Stationary Server can be connected to an available port on the Ethernet switch as show in figure 4. In this configuration, the clients do not need to complete a web authentication in order to be able to send traffic between the clients and the server, making this configuration ideal for conducting a scalability assessment of the WiFi network. Furthermore, this arrangement provides the most direct measure of the wireless networks performance because it minimizes the number of devices in the flow between client and server.
Figure 4
There may be situations where it is not possible to access the Ethernet infrastructure, commonly an issue in a small hot spot deployment, or where the service provider wishes to include the effects of the web authentication device. In this case, one is generally advised to use WaveDeploy Pro and manually navigate the web authentication function. An Stationary Server is placed onto the Internet, often in the Service Provider data center, that can be used by the technicians as shown in figure 5.
Figure 5
Once you have your topology selected, there are still a few more tasks that can be completed before you go onsite. First, make sure that the Ethernet ports are provisioned for your use prior to arrival and be sure to understand how you will know which ports to use and how to access them. Secondly, provision IP addresses for any devices that will be statically configured with IP information. Finally, have any onsite resources run a quick connectivity check between the two locations where the WiFi client and the Stationary Server are located (or as close to that configuration as you can reasonably get). There are sometimes issues in larger deployments with controller-to-controller traffic or segmentation of the Ethernet network.
13
14
SLA Example Calculation The following is an example of how to determine SLAs for a stadium deployment. Stadium deployments certainly qualify as some of the most challenging WiFi deployment environments because of the smart phone density, high user expectations, high traffic load, and challenging RF environment. It is a convenient example to use because it is easy for most readers to visualize. Nevertheless, it is important to recognize that the same techniques described in this section will also work for other venues such as concerts, fairs, and conferences. We will start with the assumption that the business team has indicated that 50% of the attendees to a given event will have WiFienabled smart phones. Base on previous events, it is assumed that 80% of the devices will be inactive, which means that they will be connected to the network, but transmitting only a tiny amount of data. The expectation is that 25% of the active users will be using voice. All of the active users will be transmitting and receiving web data at 750 kbps in each direction. This assumption is because the event coordinators noticed that a large number of people are uploading pictures and downloading a variety of material during the event. The network being designed is an 802.11g network to maximize compatibility with existing smart phones. An 802.11g access point has a maximum PHY rate of 54 Mbps, which corresponds to a maximum of about 30 Mbps of application traffic once protocol overhead is accounted for. The design team decides to target 15 Mbps of worst-case application load per each AP to provide some performance tolerance to account for likely deployment conditions such as a lower actual PHY rate, some interference, and smaller frame sizes. We can now write an equation for the total traffic load as: Total Load = TC x 0.8 x BL + TC x 0.2 x WL + TC x 0.2 x 0.25 x VL Where: o TC = Total number of client devices o BL = Background load per inactive client o WL = Web load per client o VL = Voice load per client
If we then use our formulas to derive our client counts again, we get:
o 31 Inactive Clients generating 10 kbps of background load each o 6 Active Clients generating 750 kbps upload and 750 kbps download web traffic o 2 Active Clients generating 750 kbps upload, 750 kbps download web and 160 kbps of voice Because it is assumed that 50% of the stadium attendees have smart phones, one can then calculate that the expectation is that a single AP will serve 39 x 2 = 78 people.
15
As a check, consider the required AP density for a small football stadium that holds 30,000 people. Note that 30,000 people would be considered a very small stadium and that many important venues are considerably larger (Wikipedia). According to the above calculations, it would be necessary to deploy a minimum of 30,000 / 78 = 384 APs to provide the necessary capacity to deliver on the above SLAs. This number is critical for testing the validity of the SLAs from a business sense. If the required number of APs is too high to justify economically then the definition of the project or the SLAs will need to change accordingly. This definitional work should be done prior to any deployment. If the SLAs are acceptable, it would be wise to ensure that the proposed solution can deliver the requested level of functionality in the lab prior to deploying hundreds of APs. It is preferable to use the WaveTest solution to perform this testing because it offers higher scalability and more flexibility of test configuration than the WaveDeploy solution. Once the lab testing is complete, the WaveDeploy SLAs can be configured in accordance with the information above. The target client should be configured with the maximum number of applications that people are expected to use. In our example, this would mean upstream HTTP, downstream HTTP, and voice traffic. The remainder of the clients will be created in the VeriWave hardware in the loaded network condition.
16
17
Once all of the issues are addressed then it is reasonable to proceed to the next step: concurrent application traffic with the target client as defined in the next section of this guide. It is most common for performance to be unacceptable in this initial deployment speed test. Usually some locations can provide the necessary performance while others cannot. When examining the underperforming locations, it is useful to pay special attention to which of the APs the client used to access the network. Remember that the location of the APs from the previous verification step. If the client made reasonable choices as to which AP to use, but performance was poor in spots then you likely have an installation issue relating to that one AP. If the client made poor choices for connected APs then it necessary to use a WaveDeploy Expert to test performance from a given location to the corresponding optimal AP as described below. In the case where the client made reasonable connection decisions but still experienced poor performance, it makes sense to go back to the problem location and rerun the test while using the WF1101 from the WaveDeploy Expert solution to simultaneously capture the traffic. The WF1101 captures all of the traffic from all devices and provides additional metrics that are not available from a traditional laptop capture. Once the capture file is opened, the first thing to check for is low PHY rates and high rates of retransmission. If PHY rates are consistently low for the AP or client, for example 11 Mbps or lower for an 802.11g network then this generally indicates that there is an RF problem. The most likely source of problem, since interference has been minimized for this test, is AP placement. Retesting the point with different AP and client configurations can be used to converge on an ideal placement. In the case where the client made poor connection decisions, the tester has the choice of retaking the data using a different laptop or the WF1101. The WF1101 is capable of achieving full rate 802.11 transmission and reception without any loss or buffering. As such, it is a good solution for seeing the best possible performance the network can achieve since, unlike most clients, it can outperform any networking device. The WF1101 enables the user to control which AP it uses to make its connection to the network. Thus, the user can force the WF1101 to test any specific AP in the initial deployment. This is a bit more work for the user because the user must decide which AP to use for each location, but it provides the user with unmatched control. In addition, the capture functionality of the WF1101 is also active while these tests are running. As a result, the user is able to connect to the intended AP, perform a measurement, and then examine the capture data as described above if the performance is lower than expected. The speed test allows the user to identify a whole class of additional issues that were invisible in the previous testing: Poor AP location affecting data performance Marginal APs APs with bad radios are unable to reliably transmit or receive at rate Capacity holes Areas where additional APs are required to achieve necessary performance Note that the emphasis at this step is determining coverage via the ability of the network to deliver the necessary data forwarding performance. If a configuration cannot be identified that delivers the necessary SLA then it is worthwhile to rework the SLA expectations based upon the measured data. This method identifies issues with AP placement and operation based on capacity and not radio coverage and therefore is much more closely tied to the actual customer experience.
18
19
20
Process
During the build out phase, it is appropriate to apply the lessons learned from the previous section to select better AP installation locations and network configuration. Once the network is installed and configured, the testing cycle will be much shorter than the testing of the initial section. The primary goal of this testing is to verify that the APs are placed in the correct locations and that there are no significant problems with the placements as they relate to user QoE.
Verify AP Placement
As with the initial section, the first order of business is to run a quick passive assessment to ensure that the APs are located and configured as expected. It is common to find issues in this step from section to section as a result of communication issues inherent with a large number of people. Proper installation cannot be taken for granted because different installers often work in different sections and they do not immediately benefit from their peers experiences. Once again, make sure the AP locations and information on the maps is correct and as expected before proceeding further.
21
22
Apply Process
Another major issue that we see with the shift from delivering a best effort, hospitality service to a high availability, constant quality service is that the process for resolving issues found during deployment testing in the field occurs in a very ad hoc manner. Multiple elements are often changed at the same time and retesting proceeds in a haphazard manner. This approach simply does not work because the multiple, simultaneous changes rarely converge on a viable solution. It is critical to establish a process whereby the deployment itself is done incrementally and any modifications to the installment or configuration are made one at a time and retested in order to determine if they provide the intended improvement. Tracking this work and documenting these changes is critical to identifying opportunities to improve and shorten future deployments.
23
CONCLUSION
Cellular offload onto WiFi requires a new approach to deployments which focuses on customer satisfaction with their applications in a realistically loaded environment. Adopting the best practices contained in this document will allow service providers to dramatically improve the quality of their new deployments, resulting in happier customers. With WiFi now inexorably tied into cellular contracts and the rapid rise in WiFi-enabled devices, service providers should be able to improve customer attraction rates and reduce churn by delivering a satisfying customer experience over WiFi. Ultimately, improved WiFi will generate more revenue and profit for service providers. The solution required to deliver this experience exists today. Adopting the VeriWave solutions, including WaveTest for the lab and WaveDeploy for the field, along with improving the sign off process enables service providers to significantly improve their wireless offerings.
24
WORK SITED
Goldman, D. (2011, Feb 18). Why Wi-Fi sucked at Mobile World Congress. Retrieved Feb 28, 2011, from CNNMoney.com: http://money.cnn.com/2011/02/18/technology/mwc_wifi/index.htm Phifer, L. (2010, May 18). VeriWaves WaveDeploy Raises the Bar on WLAN Assessment. Retrieved Feb 24, 2011, from Wi-Fi Planet: http://www.wi-fiplanet.com/news/article.php/3882671/Veriwaves-WaveDeploy-Raises-the-Bar-on-WLAN-Assessment.htm Wikipedia. (n.d.). List of American Football Stadiums by Capacity. Retrieved Feb. 15, 2011, from Wikipedia: http://en.wikipedia.org/wiki/List_of_American_football_stadiums_by_capacity
25
APPENDIX A: CORRECTIVE ACTIONS RESULTING FROM ADOPTING CELLULAR OFFLOAD SITE DEPLOYMENT BEST PRACTICES
This section summarizes the problem issues that are commonly present in deployments. The table highlights the problem, and the means by which it is identified using VeriWave.
Problem
Identification Technique
During a passive assessment, the AP does not show up in the list of detected APs. During the speed test the AP underperforms. Examination of the data shows a high level of 802.11 retransmits and low PHY rates in the associated traffic. Testing from a line of sight to the AP results in high performance and improved 802.11 layer functioning. Physical examination of the AP may indicate a poor installation choice. During the speed test and the ecosystem test, the system is unable to deliver acceptable QoE in locations far from existing APs. This will commonly occur near the edges of the served area, but may also occur within a served area where no AP sufficiently services an interior location. During the testing, it is noticed that they AP is set to the wrong channel or power level with respect to the planned installation.
Typical Solution
Verify that the AP is properly configured Replace the AP if it still doesnt work
Non-functional AP
Insufficient coverage
Move the APs as necessary if it appears that they can serve the existing coverage area, but are simply located poorly. Add more APs if the existing APs are insufficient to cover the necessary performance areas
Improper AP configuration
The speed test reveals that the network is not delivering as much capacity as it should in comparison with the lab testing. Checking the 802.11 PHY rates and retransmission does not indicate any issues with the PHY layer. Examining the traffic using the WF1101 in capture mode does not indicate interference or excessive management traffic as a source of the problem.
Rerun the test using the WF1101 to determine if the network of the client is the source of the problem. Check the configuration of the wired network and controllers if the network is at fault. Test to different locations if possible to identify the location of the limit. Make sure the proper firmware is loaded onto the devices. Upgrade firmware as necessary or appropriate. If testing to a central site in the Internet, consider installing an EF1101 at the local site and retesting. If the expected performance is achieved in the local network then the problem is the Wan backhaul.
26
Problem
Identification Technique
Typical Solution
If the number of beacons on the channel are very high, consider lengthening the beacon interval to reduce the amount of overhead traffic, making more traffic available for data traffic. If there is a high level of interference from external networks then consider changing the channel allocation model of the AP and its neighbors to reduce the cochannel interference with the external source. Remove any rogue APs from the facility. Mitigate non-802.11 sources of interference by selecting alternative devices that run in different frequency bands, disabling unnecessary devices, or changing channels on related APs to avoid the interference. Disable the lower PHY rates so that legacy clients or clients far from the respective AP do not consume disproportionate amount of the medium. Lower the AP power to reduce the served area and to force clients to roam sooner than they otherwise would. Consider implementing QoS in the network if appropriate. Reduce the number of clients and traffic served by an AP. Try a different client type to see if the client has an inherent issue. These results can and should be tested in the lab first and then in the field. For very high value clients, a separate SSID can be deployed to provide dedicated resources or client configuration changes can be applied. An example of this type of device would be a handheld ticket scanner at a large venus.
The speed test reveals that the network is not delivering the expected capacity. Inspection of the traffic using the WF1101 WiFi device reveals a substantial amount of additional traffic or interference.
In the WaveDeploy ecosystem test, the application SLAs will show red even though the aggregate throughput is as expected.
Traffic levels are met, but the SLA for certain application types are not met.
27
Only VeriWave Tests wireless and wired LANs as one single autonomous network Delivers real-world QoE measurements from the point of view of the end-user Creates test traffic that mimics vertical deployments such as healthcare and education Offers visibility into future network behavior through what-if test scenarios
28