Two Platforms, One Marvis
HPE said at Investor Day that Central and Mist stay two platforms sharing one AI engine. Here's where that stack stands in October, what I'm using on my own tenants today, what I'm waiting on that would help my customers, and the tools I can build around it now without waiting for anything.
The Minimum Rate Box Moves the Beacon Too
The standard says a beacon goes out at a basic rate, and every vendor picks the lowest one. Central has a box that says minimum transmit rate. I had an AP, a sniffer and a question about whether those two sentences are about the same thing. They are, and the proof is four hex bytes long.
Wi-Fi 7 rollout
Wi-Fi 7 MLO Isn't a Speed Feature. Walk Around and You'll See Why.
Same phone, same AP-735, same walk through the house. With MLO on the download averaged 430 Mbit/s and with it off 403. What actually changed was reconnections: none in 16 minutes with MLO on (one in a later run), 13 in 16 minutes with it off.
Wireless Academy
- 1What a Radio Actually Sends
- 2Bands, Channels and Widths
- 3The Link Budget
- 4Modulation and Data Rates
- 5Airtime Is the Only Resource
- 6Interference From Yourself
- 7Noise, SNR, and Why RSSI Lies
- 8Joining a Network
- 9Roaming: the Client Decides
- 10Capacity, Not Coverage
- 11Capacity, Not Coverage
- 12Surveys and What a Heatmap Can't See
- 13A Troubleshooting Method
A Troubleshooting Method
Every Wi-Fi ticket lives in one of four layers, and every layer has a counter that proves it or clears it. This is the method the other eleven lessons were building, and a lab that breaks one AP three ways so you can practice finding each with the right tool.
Your Wi-Fi Takes Turns, and the Slowest Talker Sets the Pace
Every time a Wi-Fi radio wants the air, it waits for quiet and counts down a random number, and every radio on the channel plays the same game. A laptop linked at 1201 Mb/s that sends one packet per turn moves 55 Mb/s, and one old laptop downloading beside it can take most of the air.
Surveys and What a Heatmap Can't See
A heatmap is a record of what one instrument heard, on the channels it was told to listen to, on the day. What each kind of survey proves, what the picture leaves out, and the habit of reading the AP's real channel from the controller before you believe the map.
Capacity, Not Coverage
Coverage asks whether everyone can hear the AP. Capacity asks whether everyone gets a turn. The cell isn't drawn by power, it's drawn by the slowest rate you allow, and raising that rate is the cheapest capacity you'll ever buy.
Capacity, Not Coverage
Coverage asks whether everyone can hear the AP. Capacity asks whether everyone gets a turn. The cell isn't drawn by power, it's drawn by the slowest rate you allow, and raising that rate is the cheapest capacity you'll ever buy.
Roaming: the Client Decides
The AP never moves a client. It can tell the client who its neighbors are, suggest a better one, or throw it off and hope. What the client is actually doing when it decides, what 802.11k, v and r each change, and a walk test to watch it happen on Mist and Central.
Joining a Network
A client doesn't connect. It climbs a ladder of a dozen exchanges, and every rung is somewhere it can fall off. This is the ladder in order, what each platform shows you for each rung, and a lab that breaks it three ways so you learn which rung each failure lands on.
Your WPA3 Transition SSID Didn't Break the Printers. PMF Did.
Flip a Personal SSID from WPA2 to WPA2/WPA3 transition and a class of older clients falls off, with the same name and the same passphrase. One bit and a longer AKM list did it, and the fix is a design decision, not a driver hunt.
Noise, SNR, and Why RSSI Lies
A client at -67 dBm can be flying or crawling and the RSSI number won't tell you which. The noise floor will. Where the floor comes from, why a Wi-Fi card can't see most of what raises it, and a lab where you raise it yourself with a microwave oven.
Interference From Yourself
A reader with twelve APs on one big floor asked whether adding APs makes the edge worse, and whether anything self-tunes. The answer is that AP count was never the variable. Channel width is, and the automation only moves the two knobs you let it.
Airtime Is the Only Resource
Before a Wi-Fi radio starts a transmission, it waits for quiet, draws a random number and counts it down, and every other radio on the channel plays the same game. A laptop linked at 1201 Mb/s that sends one packet per turn moves 55 Mb/s. In the site's model, one old 802.11a/g laptop downloading beside it takes its download from about 720 to about 480.
Modulation and Data Rates
The speed of a Wi-Fi link isn't a number, it's a row in a table, and the radio picks a row for every transmission. Neither Central nor Mist will tell you which row. This is how to read it backwards from what they do show, then cap it on purpose and measure what it cost.
A RADIUS Timeout Means Somebody Dropped It, Not That It Never Left
A 2930 timed out on every EAP-TLS auth while the APs at the same site worked, and the firewall log showed a fixed source port of 1812. The port was a red herring. The server was silently dropping requests from a source it didn't trust.
Your Captive Portal Works, Phones Just Won't Tell You Why It Doesn't
Laptops pass a ClearPass Guest login and phones sit in pre-auth. The page is fine; the phone's popup rejects the AP or gateway's portal cert on the post back, and one openssl command plus Access Tracker shows which of three places it died.
The Link Budget
Four numbers tell you how loud your signal will be when it arrives. I wrote the budget for my own lab, walked thirty-two feet, and came up fifteen decibels short with nothing in the way. The gap was real. It was hiding in the one term everybody sets to zero.
Your AP Is Not Moving That Client
Roaming is a client decision, band steering is a suggestion, and the phone that sits disconnected after an elevator is running a documented backoff timer. Opening the Wi-Fi settings screen does not wake it up, and the popular explanation of why it seems to is mechanically wrong.
A Captive Portal Page Name Long Enough to Reboot Your APs
A 605H in bridge mode kernel panicked every two to three minutes and nothing in the WLAN config looked wrong. The cause was the length of the captive portal page name, not the page, not the policy, not the AP. Here is why bridge mode is what exposed it, and why tunnel mode would have hidden it.
ClearPass, properly
Which ClearPass Node Is Actually Polling Your Switch
SNMP reads do not come from the publisher. They come from whichever node the cluster decides to use, which matters the moment you write a firewall rule or a control plane ACL that names one address. Here is how to prove which node is sending, and why ARP polling is the only thing that classifies a static device behind a firewall before it authenticates.
Bands, Channels and Widths
A band is a stretch of spectrum, a channel is a slice of it, and width is how big a slice you take. Every doubling of width doubles the rate, halves how many cells you can have, and costs 3 dB. There is a game in here to prove the middle one.
Migrating APs
Converting a Mesh Over the Air Strands Every Point
HPE says the mesh profile survives conversion, and the same doc set says a mesh must be provisioned over the wire the first time. Nobody has reconciled those two sentences, and the one that is wrong is the one that costs you a lift. Plan for the wire.
The Wizard Is Lying About the Shape of Your Config
A tunneled MPSK client that authenticates and lands nowhere, a Central NAC certificate that never gets issued, and a device quietly overriding group policy. Three features, one shape, and none of them raise an error.
The Simulator Is a Wi-Fi Link You Can Break
The simulator on this site sends a real frame from a packet capture, one symbol per particle, through a link budget you can drag walls into. Four minutes on what it is doing and why the Teams call breaks before the chat does.
What a Radio Actually Sends
Frequency, wavelength, and the dBm math you can do in your head. Then a lab where you walk away from an AP and watch the number fall exactly the way the formula says it will.
Your Survey Tool Can't See Channel 173
A client sat on one AP for 49 minutes while the signal fell to -69 dBm, and every heatmap said coverage was fine. The radio it was stuck to was on a channel the survey never scanned.
Your ClearPass Intune Integration Probably Authenticates Nothing
Most ClearPass and Intune builds look up a MAC address and call the result authentication. Here's the version where the certificate does the authenticating, Intune does the authorizing, and the two are tied together by something the client can't lie about.
One Pilot AP Converts All Hundred of Them
Someone with a hundred Instant APs on AirWave wanted to move them to a controller cluster and, sensibly, test on one AP first. The virtual controller had other plans.
Your ClearPass SAML Login Validates and Still Fails
Admin MFA on ClearPass means SAML to your IdP, and every guide you find is about Onboard instead. Here are the three places an admin login fails silently, in the order to check them.
Switching, carefully
Your VSX Upgrade Is Hitless Right Up Until It Isn't
Everybody has heard a VSX upgrade is hitless. Almost nobody has been told what hitless is conditional on. It's a single cable and a 180 second timer.
When the Docs Disagree With Themselves, Somebody Has to Decide
Three official minimum versions, a withdrawn release note, and a command renamed without a memo. How to get to a decision you can put in a change record.
84 Percent of Your Airtime Is Printers Saying Hello
Someone counted. 269,253 mDNS frames in under six minutes, 84 percent of an uplink, and every one of them going out over the air at 6 Mbps. Here's the math nobody does.
AOS-8 to AOS-10 Is a Rebuild, Not an Upgrade
Three threads in one week, three symptoms, one cause: people planned an upgrade and got handed a rebuild. What doesn't come with you, the pre-validate defect everyone blames on DNS, and the order of operations.
Your New AP Isn't Broken, It's Hungry: PoE Budgets in the Wi-Fi 7 Era
Wi-Fi 7 APs don't fail when they're underpowered. They negotiate, shrug, and come up in a reduced mode: USB off, second port off, 2x2 instead of 4x4. Nothing turns red.
Field notes, not a newsletter
Campus networks, wireless, NAC, and the forum threads that keep asking the same question.
