CWNE Essay #1
Keeping the Robots Happy
A wireless equipment manufacturer brought me into a project after the VAR that initially won the bid failed to meet the customer's requirements. The winning VAR was a major online reseller and would not agree to any onsite post-installation validation of the wireless network.
As it turns out, the customer had some specific and sometimes incongruent project requirements. One requirement on their list was to "Measure signal strength (RSSI) in all areas for better than 65 dBm, and SNR less than 25 dB separately..." While -65 dBm is an acceptable target, 'less than 25 dB' SNR shouldn't be. Other requirements seemed to indicate someone did an internet search for "wireless network requirements" and patched together the results.
The customer specified using BYOD devices to measure RSSI, SNR, secondary coverage, latency, and jitter. While I am all for the customer testing their wireless network in whatever manner they wish, I could not allow this to be a requirement for the project. We would never be able to agree on what constitutes a successful implementation. I needed to help this customer rewrite their project requirements before accepting the job.
Fortunately, the customer was open to discussion, ultimately, they listened to and accepted advice. I was able to provide a project acceptance testing methodology and a concrete statement of work and proceeded with the project. What I didn't know at the time was that the IT department didn't have access to all of the wireless network requirements. More on this later.
The site was a multi-floor scientific research lab. The buildings were modular sections of reinforced concrete interspersed with outdoor atrium "cutouts." This made the space a maze of hallways with medium to large labs and quite a few small offices interspersed, with very few contiguous open spaces.
I started by performing a passive survey of the existing wireless network. Along with an RF analysis survey of the space, I documented the wired LAN for PoE switch port location and availability to determine cable runs. I decided to do a few AP on a stick surveys for some particularly interesting locations. After modeling and remediating the entire site in Ekahau, I provided their installation contractors with detailed installation instructions. This documentation is very specific regarding AP placement and what the installation contractor must do if they cannot use a particular location. This documentation came in handy on this project. While performing the post-installation inspection, I found nine access points installed in un-approved locations. The installer did move them when requested but attempted to bill for the moves. Due to the clarity of the installation documentation, they were not able to.
Post project testing included another site-wide passive 802.11 and RF survey, ten separate static location Jitter and Latency tests using an iPerf server on the LAN. The documentation provided.
to the customer demonstrated that all acceptance requirements had been met. Project complete… or was it?
The following issue didn't initially present itself as a wireless trouble ticket. It came up in conversation with this group of scientists while on break in the cafeteria. I overheard them discussing a problem with their robots and asked if I could look at their data.
As it turned out, one of the research departments was developing telepresence robots. Before this wireless network project, the robots only worked in one particular pod where they had their wireless network. One of this project's stated goals was to remove all of these "rogue" access points and get the whole facility on the same infrastructure. When this department decommissioned its wireless network and moved to the new one, they were initially pleased with the performance until the robots started venturing to areas outside their pod.
As the robots turned a particular corner, they would freeze for some time before becoming operational again. As the scientists investigated these freezes, they found that the robots would occasionally get stuck in other seemingly random locations as well. Since they were a group of research scientists, they kept precise data about where and when these freezes occurred. The freezes lasted anywhere from 20 seconds to almost two minutes. Then the robot would be back in operation. Because they eventually regained control of the robot, the scientists initially started troubleshooting their robotic software and hardware to find out what was happening.
As robotics scientists, they thought their problem was somehow with the robot. As a wireless engineer, I thought the issue might be related to the new wireless network implementation.
I used Profiler on the WLAN Pi to determine the 802.11 characteristics of the robot NIC's; there weren't any conflicts with the new WLAN. I looked at the RSSI and SNR information that the wireless LAN controller derived from the robot clients. The values, when connected, were within acceptable limits. It took some digging into RRM and system logs to determine that each time a robot would freeze, the AP providing primary coverage in the area was on a DFS channel. Due to the nature of the space, there were often no other associated client devices on these hallway radios. With no active clients on these radios and the robot's inability to directly probe DFS channels, roaming was adversely affected.
I reconfigured the channel plan to ensure primary coverage from non-DFS channels in the areas where the robots operate. Now they have a swarm of happy robots. Always a good idea to keep the robots happy, now if I could only get these guys to program their robots to do wireless site surveys.
Following the CWNP methodology to Define, Design, Deploy, Validate & Optimize made this project a real success. In this case, I helped the customer redefine the project requirements to create realistic and measurable acceptance requirements. The optimize step allowed me to make changes to the configuration after determining that DFS channels hindered the robots from roaming freely. Fixing the robot roaming issue was a significant win as far as the customer was concerned.