Fabric Site support for 6000+ Access Points

[packet2020]

Hi All,

I've posted here a couple of times over the past few days. Thank you for the support and guidance. Another question.

I understand from reading various Cisco publications that a fabric site can support upto 10k access points, but I was wondering how this is acheived in practice? Seeing that the largest WLCs 9800-80s and CW9800H1/H2s support a maxium of 6K access points, can multiple WLCs be associated to a fabric site to acheive a scale of 10K access points? Also to support this number, I'm assuming that a single /18 access point pool will need to be allocated to the fabric site?

Thanks

[Leo Laohoo]

---

@packet2020 wrote:
I understand from reading various Cisco publications that a fabric site can support upto 10k access points, but I was wondering how this is acheived in practice? Seeing that the largest WLCs 9800-80s and CW9800H1/H2s support a maxium of 6K access points, can multiple WLCs be associated to a fabric site to acheive a scale of 10K access points? Also to support this number, I'm assuming that a single /18 access point pool will need to be allocated to the fabric site?

---

With the introduction of the "80% Rule", it all boils down to two things: 

• Stability of the code
• WLC uptime. 

Overloading (above 50%) a multi-WNCD controller like the 9800-40/-80/-X is taking a big risk.  And I would never recommend anyone scale their controller to >90% unless they have High Touch Technical Support and the "eyes and ears" of the WNBU beside them.  

[packet2020]

Thanks Leo - If this is the case, and a best practise/rule documented by Cisco, then I assume that to acheive a scale of 10k access points per SD-Access fabric site, a minimum of three controllers (or six controllers deployed in three HA/SSO pairs) will be nessasary so that the access points can be balanced across WLCs without exceeding this rule. Unfortunatley I cant find any documenation or guidlines regading multiple WLC support per fabric site which is what I need to confirm to support this.

[Leo Laohoo]

For 10k APs and applying the "80% Rule", then yes, a minimum of 3 WLC.  

I would not even recommend HA SSO because it will not scale -- HA SSO will only cause more instability.  

Just this year, we tore apart 2 x VSS pairs of 9800-80 and coverted all of them to N + 1.  The memory leak has slowed and the in-between proactive reboots are getting longer.  

Once the control-plane memory utilization goes >40%, prepare to intervene.  Cold reboot is the best but warm reboot is better than a crash.

Take note of the following: 

• Regular reboot of the WLC:  Once every four to six months if <50% AP load.  If the AP load is about 80%, then one every 2 months. 
• AP Count:  <50% of what is stated in the Data Sheet
• Client Count:  <50% of what is stated in the Data Sheet
• Regular reboot of the APs:  Daily or weekly
• Inter-controller roaming:  Zero
• Mobility Group:  Minimize to ZERO
• PSK:  Preferred over anything else or it will stress out the WNCD
• IF there is no DNAC, turn off "telemetry" &/or "NMSPD".  
• Web Authentication:  Don't.  Just don't.  Use PSK for public WiFi and avoid Web Authentication.

[jedolphi]

Hi @packet2020 , a few more notes FYI:

• Yes, only one AP pool per fabric site, so you need to make it large enough to work on day 1 and accomodate possible future wireless growth. Since there is no layer 2 flooding in SDA by default, and never any layer 2 flooding in the AP management VLAN, it is perfectly safe to make it a single large subnet with 10K IP addresses (10K APs) in it.
• We can have multiple WLCs or SSO WLCs in a fabric site. When you provision a WLC to SDA fabric you are asked what area the WLC is managing and what area the WLC is seconday (N+1 backup). WLC1 might manage building 1-5, WLC2 might manage building 6-10, and so on.
• Theoretically there is no limit to number of WLCs in a Fabric Site, although I'd be very surprised if anyone exceed 8, there is no reason to do such a thing.
• CatC will automatically setup mobility tunnels between multiple WLCs in same SDA Fabric Site.
• Please re-check CatC data sheet, it can do up to 25K APs in a Fabric Site with 3x XL CatC cluster. Please don't design to this enormous scale though without first talking to your Cisco presales rep. Something that large is very unusual and should have a second set of eyes to make sure it doesn't cross any other boundry e.g. max concurrent endpoints, max transient endpoints, etc.

Best regards, Jerome