ArubaOS 10 / controller LCM

Nomenclature note: in ArubaOS 10, "MDs" become "mobility gateways" or simply "gateways" for short.

AOS 8 requirements:

At a minimum, we need ten (10) 9240s on campus.

  • 9240s (with a gold license) can handle the same number of APs/clients as 7240s on AOS 8.
    • 2k APs
    • 32k clients
  • With 8800 APs on campus, we at a minimum 5 controllers online for all the APs to be up. Double that for all APs to have a standby tunnel built.

AOS 10 requirements:

We need 8 total 9240s.

  • Campus: 4x (gold license)
  • equinix: 2x (base hardware)
  • vtc: 2x (base hardware)

Reasoning:

  • A single 9240 (with gold license) can handle:
    • 16k APs
    • 64k clients
  • With 2 gateways in each switchroom (bur/col), we can lose a whole switch room (power, meteor, whatever) and have all the APs still be up with a redundant connection.
  • Why not 1 gateway per switchroom?
    • client count would be tight in a failure scenario
    • a failure of any kind leaves us without any redundancy
    • we'll already have the hardware (which is relatively cheap to begin with) due to AOS 8 requirements
  • For off campus locations (equinix/vtc), we just need redundancy.

Migration plan

Phase 1 (AOS 8):

  • License them all with a gold license
  • Deploy them on EVPN
    • bur: 5
    • col: 5
    • cluster configured with two groups, so each AP has their AAC/S-AAC to distinct switch rooms.
  • Move campus APs to the new 10 node cluster
  • Use campus 7240XMs for dev/pprd
    • probably leave the col/bur MDs in place for phase 2 (though we will need to move fiber to connect them to EVPN)

Phase 2 (start moving APs to AOS 10):

  • Setup at least 4 7240XMs on EVPN with AOS 10
    • probably 2 each in bur/col
    • exact number and placement subject to future consideration
  • Migrate at least 3k APs to AOS 10 cluster
    • This will certainly be done in stages and after lots of dev/testing, etc
    • We need few enough APs on the AOS 8 cluster to be able to remove 4 controllers. This leaves us with 6x 9240s, which can handle 6k APs.

Phase 3 (campus 9240s to AOS 10):

  • Remove 4 9240s from the AOS 8 cluster and add them to the AOS 10 cluster
  • Remove 7240XMs from AOS 10 cluster
  • Move the rest of campus to the new AOS 10 cluster
  • Optionally consolidate fiber links so the gateways are connected at 4x 25G
    • We should have plenty of links left over from consolidating the cluster from 10 MDs to 4 gateways.

Phase 4 (remote locations):

  • Downgrade license on remaining 4 9240s to base hardware
  • Use 4 now empty 9240s for VTC and equinix (AOS 10)
  • Use remaining 2 9240s in dev

Bonus Phase (stadium):

  • Use dev 9240s in the stadium (upgrade license to gold)
  • Buy 9000 series (or use VMs) for dev/pprd