Dot11 Guru

802.11, Design, random ramble

A unique opportunity – CWNE Essay 1

Last year, I was asked to visit a site to refresh and redesign their wireless network.  This was unique because the initial design was also mine, it was the first wireless design I’d ever completed.  What followed was a period of reflection as I reviewed the existing design, and noted on the mistakes I’d made.  The following summarises what I’d learned over the 5-year journey in terms of 802.11 design principles. 

  Knowing your clients

There was a time where the focus on wireless deployments was predominantly on the infrastructure, and some rudimental RF metrics (RSSI, SNR,).  I quickly learned that every wireless client performs differently on the network, and it’s important to know your clients.  This became very apparent when I started designing wireless networks for robots and automation in our factories and warehouses. 

I built a database which contains a list of our clients, RF design considerations and capabilities provided by the vendors, and the offset between our survey equipment and critical clients.  This database is then used to help determine the RF metrics needed for each individual site.  If a site adds a new device which wireless is considered critical, they are required to notify me so that we can review the requirements before implementation. 

Access Point (AP) placement 

This site is a typical enterprise office, and over the past 5 years, the reliance on wireless had dramatically shifted.  That said, I noticed a big change in the way I placed APs.  Initially I had placed the APs so that they would cover as much floor space as possible.  I also made the mistake of placing APs in the exact same position on each floor.   

During the redesign I removed omni-directional APs from hallways and placed them into the meeting rooms.  I placed access points closer to the users and changed power settings to 25mW.  This was done to help match the power in which clients transmit back, and also to contain BSS allowing a smaller number of clients on each access point.  I also ensured APs were not placed in the same position on each floor. 

The change in design and power settings has increased the amount of APs we would install into a site.  This combined with the power requirements from the Cisco 3800 series access points has put a real strain on our POE budget.  During the deployment we encountered a number of issues where the switches would run out of POE.  Luckily this was covered in the initial design and we had enough switches to balance the load.  The issue we faced was a communication issue with the people patching the access points, where they were not told to perform load balancing across switches. 

Channel Assignment vs Co Channel Interference 

Initially the site was designed to use all 5GHz channels possible, 40MHz wide.  After a lot of research in our lab, I found that for our environments we were better off 20MHz wide, and with fewer channels used.  I now exclude DFS channels from our channel plan.  This is because of the long roam times associated with AP discovery and scanning times required by regulation on these channels. 

I am also trialling a method where we accept a little more Co Channel Interference (CCI), and only utilise UNII1 channels.  This is because of a suggestion from a voice vendor at the WLPC conference in Prague.  The suggestion is that the channel scanning for 21 channels consumes too much time, and that it’s better for the voice client to scan less channels and compensate with more CCI. 

Static Assignment vs Radio Resource Management (RRM) 

I am now responsible for sites in the Asia Pacific, China, Europe, Middle East, and Africa sites.  I’m not alone, I have a number of teams who help a lot, but I am the final escalation point for issues and decisions.  This means that usually, the positives of RRM far outweigh the negatives.  I have spent many hours with RRM and have defined a process for determining if RRM will work for the site, and how to tune it.  This site was no different. 

Assuming a site has enough access points to be able to use RRM I follow the following process.  Using a predictive survey, I calculate the average maximum power level of TCPv1.  This is done using the following formula   

TXIdeal=TXMax+(TCPv1Threshold−RSSI3rd) 

The Threshold stays at -70, the maximum TX is the maximum supported by the radio in the regulatory domain.  Calculating and enforcing this helps control situations where an access point may exceed the amount of power we’d ultimately like to see.  DCA is set to remove DFS channels, enforce 20MHz channel width.  The intervals are changed to 12 hours after the deployment is finalised. 

Once the configuration is completed, time must be given for the algorithms to process.  I usually allow 3 hours after forcing the calculations to begin (a good time for dinner). 

The next step is to perform an intermediate site survey.  I am only focusing on taking one or two data points in each meeting room or office.  This is to get an idea of how the co channel interference seems, and if more adjustments are required before the validation survey. 

Client Validation 

The final step of any wireless deployment is to validate the client behaves as required on the network.  This is especially important for our deployments as we are an internal team and hold some responsibility for end to end performance of each application.  This testing is always done with the customer.  I perform the validation in this way so that our customers can see the working product, and if a problem does occur for an expected reason, I can educate them on why this is the case.  For an enterprise office such as this, the test involves a cabled user playing music on a voice call with a wireless user who is roaming amongst the office (and vice versa).  We are listening for major packet loss and unacceptable roam times.  

Reflection is great, and I can see I have come a long way in the time I’ve spent focused on wireless design.  I am sure if, given the opportunity to re-visit in the future, I will see where I have improved from today.  The initial project was deployed using Air Magnet, most recently we have switched to Ekahau. 

Share this:

Leave a Reply