With wireless now the preferred, default, and increasingly only access in the majority of in-building, campus, metro-scale hotspot and wide-area settings, achieving optimal performance is a key objective for IT departments.\nSince radio-frequency (RF) propagation always involves a high degree of variability, it\u2019s often difficult to predict the precise behavior of a given installation. Variables include operating conditions, user and application traffic demands, and the capabilities and settings of individual vendor products. When mobility, Wi-Fi testing and verification are also taken into consideration, performance evaluation can become very complex indeed.\n\nPerformance variables include throughput, coverage, time-bounded traffic (primarily telephony and streaming video), reliability, security, rate-vs.-range behavior and traffic varying with motion, range, time of day and location. Perhaps the most important is capacity \u2013 the ability to successfully address the requirements of all traffic demands in any given location and at any given moment.\nWith Wi-Fi equipment price\/performance ratios continually improving with advances in standards, technologies and implementations, many IT shops have taken a brute-force approach: Just upgrade and\/or add access points (AP) and Wi-Fi controllers, Ethernet switches and related hardware as indicated.\nUnfortunately, this approach often doesn\u2019t optimize performance and is costly in terms of both equipment and the time required of network operations professionals.\nIn the interest of finding a better way, we interviewed a number of experts to derive a set of best practices for Wi-Fi performance optimization. We focused on the three major phases of any large-scale Wi-Fi installation: planning and pre-installation; post-installation functional and performance testing and verification; and dealing with ad-hoc performance issues (troubleshooting). We also sought suggestions for a requisite operations toolset, and, we asked about future directions for Wi-Fi performance optimization.\nOur interviewees include Dr. Eldad Perahia, distinguished technologist in Aruba\u2019s CTO Office; Matthew MacPherson, senior director, WLAN, Jim Florwick, senior technical market engineer, and Nilesh Doshi, software test manager, all from Cisco; Ekahau CEO Mika Hakala and product manager Jerry Olla; Mike Leibovitz, senior director of product management at Extreme Networks; Ali Youssef, principal mobility architect at Henry Ford Health System; Bob Friday, co-founder and CTO of Mist Systems; and\u00a0Dr. Leigh Chinitz, CTO at octoScope.\nWi-Fi performance optimization: planning stage\nOur experts agree that much can be done to optimize solution performance and to prevent operational issues, even before deciding upon specific equipment and its subsequent installation.\nHakala recommends designing for end-user quality experience (QoE), capacity and latency, rather than coverage alone. Olla adds that understanding user location, density and application demands are key to assuring capacity. Olla also notes that surveying physical space to understand RF behavior can avoid numerous opportunities for headaches later.\nChinitz suggests that testing individual access points under realistic operating conditions and loads can assist in both understanding the limitations of individual APs as well as in the calculation of capacity. Adding a second AP enables the testing of roaming, load balancing and bandsteering, and can in most cases reveal location-specific path loss and dead zones.\nYoussef says a walk-around \u201cAP-on-a-stick\u201d test can be useful in potentially difficult environments, like hospitals. He\u2019s even seen interior paint negatively impact RF performance.\nMist Systems is a major proponent of bringing indoor Bluetooth Low Energy (BLE)-based location on par with outdoor GPS. Friday argues that the higher-density AP spacing ideal for indoor use cases is also beneficial in optimizing Wi-Fi deployments.\nFlorwick recommends an initial RF sweep for potential interferers in the unlicensed bands, Wi-Fi or otherwise, and to pick initial Wi-Fi channels accordingly. Optimal channel bandwidth is often a function of both interference and channel utilization.\nPerahia suggests that building analytical models of the anticipated load, even using just a spreadsheet, can be useful in analyzing capacity assumptions.\nLeibovitz says that a careful examination of the remainder of the network (beyond Wi-Fi alone) for capacity, management, and local services like DHCP, can avoid problems down the road, as wire-side issues often reveal themselves on the wireless elements of the network. It\u2019s also important to consider emerging, potential and anticipated \u2013 not just current\u00a0\u2013 traffic demands, such as those from IoT.\nWi-Fi performance verification\nAll of our participants agree that post-installation functional testing and performance verification are essential and should include at least a minimal sample of users with production applications under realistic operating conditions.\nYoussef says that he usually sees more issues related to products, particularly client devices, than RF. Wide variability in client behavior and compatibility is common, along with issues such as driver settings inappropriately changed by otherwise well-meaning end-users. He suggests a cursory re-survey of deployed space annually, noting that facilities organizations also sometimes move or deactivate APs for other-than-nefarious purposes.\nPerahia cautions that inappropriate (out of spec) or defective AP wiring can result in unexpectedly poor performance.\nDoshi recommends alpha testing a new deployment before allowing users on, measuring such elements as voice quality and roaming behavior. MacPherson suggests using analytics tools and making use of facilities commonly available in management consoles.\nLeibovitz suggests occasional spot checks in multiple locations with a specific end-user device, even in installations that appear to be functioning normally. \u00a0\nOlla points out that documentation is critical at all stages of deployment \u2013 even observational notes can provide solid clues when verifying functional performance.\nWi-Fi troubleshooting and toolsets\nOur participants note that the software and\/or hardware toolset used to verify performance is also appropriate when troubleshooting, so it\u2019s important to have both data-gathering and analysis\/analytics capabilities that cover RF, wireless network behavior, the wired network, and in some cases the client and WAN as well.\nLeibovitz recommends that installations should gather as much operating data as possible, since it\u2019s easier to spot patterns when a lot of data is available. Analytics capabilities are increasingly important here.\nOlla suggests checking the remainder of the network value chain for problems despite user protestations that the issue is by definition \u201cwireless.\u201d Chinitz adds that talking with end-users can yield valuable insights.\nYoussef recommends the use of a third-party, standalone Wi-Fi analysis and assurance tool, independent of the selected management console, so as to get a second opinion beyond the management console and gain access to a different set of troubleshooting strategies.\nFriday says having a real-time cloud-based management stack with API access enables superior automation and troubleshooting for end-user organizations and third-party toolset providers alike, enabling new use cases to be quickly developed and deployed. He also notes the availability of natural-language query systems that can answer questions on par with a network domain expert, and expects additional speed and accuracy in these over the next couple of years.\nPerahia recommends testing particularly vexing situations with a second AP from a different vendor. If the results are uniformly poor with both, the chances are that the client is at fault. He also recommends the use of packet-sniffing software, and even simple benchmarking tools like Iperf can be useful in quickly isolating troublesome clients, APs and traffic types.\nEven after almost three decades of the widespread use of WLANs in organizations everywhere, there\u2019s clearly still much that can be done to augment performance optimization and related operational elements.\nFuture trends in Wi-Fi performance optimization\nThe Wi-Fi experts we interviewed say they expect to see more sophisticated analytics tools, including advanced capabilities based on artificial intelligence (AI) and machine learning (ML). These capabilities are already enabling the automated remediation of issues in advance of operations-staff awareness. Friday refers to these as self-driving networks.\nChinitz talks about virtual benchmarking, which involves testing an access point (and potentially even client devices) in a specially designed RF-isolated chamber under a very large set of conditions, with the results then quantified into a profile than can be used to predict real-world results.\nAccording to Perahia, future operational-performance analysis and optimization will depend upon information gathered directly from client devices and not just infrastructure. The reason for this is the increasing utilization of beamforming, multi-user MIMO, and other techniques that can result in traditional sensors and assurance tools being in an RF null during a particular transmission. WLAN chipset vendors will likely be required to augment client functionality here, so as to gather the required performance data.\nMacPherson says that SDN will likely play an increasing role in both data gathering and in optimization within the wireless segments of the organizational network. He also notes that management console interfaces will need to be simpler and more precise, and leverage the rise of AI and ML.\nAt the very least, it would be helpful to have a basic benchmarking server, perhaps something like speedtest.net, and as simple as iperf, implemented within APs, so that the actual performance of a given airlink can be analyzed easily. This is a big step up from knowing only the reported signaling rate, which can vary widely and rapidly in operation, and, regardless, is of little use in performance optimization.