Tuesday, February 10, 2015

Hallway Design Nightmares Part3: Channel Selection

I'd like to start with the fact that I am an RRM guy.  I think Radio Resource Management does a pretty good job of managing TX Power and Channel selection when configured properly.  I know that some people out there don't have confidence in RRM, just know that this reflects my opinion regarding the matter.

We covered TX power in part 2 of this series, now let's talk about Dynamic Channel Assignment (DCA).  I can't tell you how many times I've been talking with a customer or fellow engineer and they tell me "DCA is broken," "DCA makes poor choices," or "DCA is making poor channel selection choice."

When I ask them to show me why they think this, they generally show me a map that has multiple APs in the hall and several of them are on the same channel and adjacent to each other.  Much like the following:

------------------------------
|      1     1     1     11     |
------------------------------




(Apologies for the ASCII art, I lost the 3D model I built for the building from post 1 and 2)

Now at first glance, it appears the channel selection is far from optimal.  But if we look at what is really going on, let's redraw this with the floor above and below:

Top Floor
------------------------------
|      6     6     6     11     |
------------------------------
------------------------------
|      1     1     1     11     |
------------------------------
------------------------------
|      11   11   11     1     |
------------------------------
Bottom Floor

And let's look at the Nearby APs from the CLI output on the WLC.
(sorted and filtered for your reading enjoyment)

AP     00:22:00:00:00:e0     slot     0..................     -42     dBm     on     6
AP     00:23:00:00:00:b0     slot     0..................     -43     dBm     on     6
AP     00:23:00:00:00:30     slot     0..................     -44     dBm     on     11
AP     00:23:00:00:00:90     slot     0..................     -47     dBm     on     11
AP     00:23:00:00:00:c0     slot     0..................     -47     dBm     on     1
AP     00:23:00:00:00:70     slot     0..................     -47     dBm     on     1
...
Output from command:
show ap auto-rf 802.11b <AP Name>


As you can see, from the APs perspective, it is hearing the APs on the adjacent floors better than it hears the APs down the hall.  I can't comment on why from a specific example, but it was due to the hallway construction.  From the APs perspective, it did select the correct channel (1 is better than 6 or 11).

From Part 2, we turned up the power of the APs to push through the walls and get RF into the rooms, and we tried to reduce co-channel by turning off radios in the 2.4Ghz.   I've marked the APs in Red that we would have turned off to start with had this been a real building.

Top Floor
------------------------------
|      6     6     6     11     |
------------------------------
------------------------------
|      1     1     1     11     |
------------------------------
------------------------------
|      11   11   11     1     |
------------------------------

Now next day I get the call that the channels haven't changed.  Any guess as to why?

DCA Channel Sensitivity:
One of the ways that RRM prevents APs from changing channels too rapidly for minor gains is with DCA Channel Sensitivity.  What this does is tell the WLC that it must expect to achieve a specified amount of improvement before it will take action.  The default is Medium (10dB).  Since all of our APs that were turned off near the bolded AP are less than 10dB, the controller took no action.


One of the things I do when making big changes to an environment is to do the following:
Either Restart the RF Grouping, Reboot the WLC, or kick the DCA Sensitivity into High (5dB) with the first 2 being my preference.  This reduces the amount of gain in dB needed before the WLC will make a change.  Rebooting kicks the RF leader into Startup mode which will aggressively change channels at 5dB.  Shortly after doing one of these, you will start to see the channel plan normalize.  It still may not be what you want due to some of the limitations of the design, but it may very well get better.

As with TPC, turning off even half of the radios may not be enough to get the WLC to change the channel plan at Medium.  Sometimes we have to kick it in order to get the immediate change we want.

NOTE: These channel changes are disruptive to clients.  You may see increased disconnects while the WLC is changing channels, and long term you probably want to return DCA to Medium.  Plan your change windows accordingly.

Edit:  One thing I really want to say is to understand what the "Avoids" do in the DCA settings.  Each one of these things has some serious impact on the DCA process.  Things like "Avoid Cisco AP Load" can have a huge affect on the channel selection and it can change very rapidly.  Go check out: http://www.cisco.com/c/en/us/support/docs/wireless-mobility/wireless-lan-wlan/71113-rrm-new.html