CX Sandbox

The commands, one note each

What each command does, two or three examples you can type into the sandbox, what changed between AOS-CX 10.15 and 10.18, and where the sandbox fakes hardware instead of reproducing it. Every example is run through the sandbox before this page is built, and every release line comes from the command lists and show output I pulled off the Switch Simulator for each release, or from HPE's guide where the simulator has no hardware. The same notes show up beside the terminal as you type.

Basics15

configure terminalGet into configuration mode

Everything that changes the switch starts here. The prompt grows (config) and stays until you type end. Unique abbreviations work: conf t is the same thing.

In every release here, 10.15 to 10.18.

Habit: Take a checkpoint before a change you might want to undo: copy running-config checkpoint before-change.
try it in the sandbox
configure terminalhostname access-01endconf tvlan 10end

Goes withendcopy running-config checkpoint <WORD>

endBack to the # prompt

Leaves whatever context you are in and drops you at the exec prompt. exit goes up one level; end goes all the way.

In every release here, 10.15 to 10.18.

try it in the sandbox
configure terminalinterface 1/1/1endconfigure terminalvlan 10exitend

Goes withconfigure terminal

hostname <WORD>Name the switch

Sets the name in the prompt, in LLDP and in logs. Letters, digits, dots and dashes, no spaces.

In every release here, 10.15 to 10.18.

Habit: Put the location in the name. You will read it in an LLDP table at 2 a.m. someday.
try it in the sandbox
hostname access-01hostname idf2-sw1
write memorySave the running config

Copies the running configuration to the startup configuration so it survives a reboot. copy running-config startup-config does the same.

In every release here, 10.15 to 10.18.

Habit: Save after the change is proven, not before. Until then a reboot is your rollback.
try it in the sandbox
write memorycopy running-config startup-config

Goes withshow startup-config

show running-configWhat the switch is runningchanged

The whole active configuration. Add interface 1/1/5 to see one port's block. Pipe it to cut it down: | begin, | include.

In every release here, 10.15 to 10.18.

Changed: 10.16 added server priority lines and the built-in radius group to the output; 10.15 printed neither.
try it in the sandbox
show running-configshow running-config interface 1/1/5show running-config | begin "port-access role"

Goes withshow startup-configshow checkpoint

show startup-configWhat it will boot with

The saved configuration, the one a reboot loads. If it differs from the running config, something has not been saved yet, on purpose or not.

In every release here, 10.15 to 10.18.

try it in the sandbox
show startup-configcheckpoint diff startup-config running-config

Goes withwrite memoryshow running-config

show versionWhich release it runs

The AOS-CX release and build. The first two letters say the platform: ML for the 6200, FL for the 6300. The release picker in the sandbox changes what this says.

In every release here, 10.15 to 10.18.

try it in the sandbox
show versionshow version | include Version

Goes withshow system

show systemModel, serial, uptimechanged

Product name and part number, serial, base MAC, uptime and CPU. The part number is how you tell a 6200F 12-port from a 24-port when nobody wrote it down.

In every release here, 10.15 to 10.18.

Changed: 10.18 added CPU idle, user and system lines; 10.17 printed CPU utilization only.
try it in the sandbox
show systemshow system | include Product
no pageStop the more prompt

Turns off paging for this session, so long output scrolls instead of stopping every screen. Every script and screen capture should start with it.

In every release here, 10.15 to 10.18.

try it in the sandbox
no pagepage 40
copy running-config checkpoint <WORD>A named restore point

Saves the running config under a name. If the running config is identical to a checkpoint that exists, 10.18 refuses with An identical checkpoint already exists.

In every release here, 10.15 to 10.18.

Habit: The box also makes system checkpoints (CPC...) after changes on its own. Name yours so you can find them.
try it in the sandbox
copy running-config checkpoint before-changeshow checkpoint

Goes withcheckpoint rollback <WORD>show checkpoint

checkpoint rollback <WORD>Go back to a checkpoint

Replaces the running config with the checkpoint. It does not ask. checkpoint diff checkpoint <name> running-config shows the difference first, and that is worth doing.

In every release here, 10.15 to 10.18.

try it in the sandbox
copy running-config checkpoint before-changecheckpoint rollback before-changecheckpoint rollback startup-config

Goes withcopy running-config checkpoint <WORD>show checkpoint

show checkpointList the checkpoints

Every checkpoint, newest first, with the startup config among them by date. show checkpoint <name> prints one. There is no show checkpoint list on 10.18: the box reads list as a checkpoint name.

In every release here, 10.15 to 10.18.

try it in the sandbox
show checkpointshow checkpoint before-change
checkpoint diff <from> <to>What changed between two configs

Compares any two of running-config, startup-config and checkpoint <name> as a unified diff: a line starting with a plus is only in the second, one starting with a minus only in the first. No difference in configs. means they match. Worth running before every rollback.

In every release here, 10.15 to 10.18.

Habit: checkpoint diff startup-config running-config before you walk away tells you whether anything is unsaved.
try it in the sandbox
copy running-config checkpoint before-changecheckpoint diff checkpoint before-change running-configcheckpoint diff startup-config running-config

Goes withcheckpoint rollback <WORD>show checkpoint

checkpoint auto confirmKeep the change

Ends a checkpoint auto with the change in place. I didn't capture what the box prints for it; the sandbox prints nothing. show checkpoint keeps auto-checkpoint-start-backup either way.

In every release here, 10.15 to 10.18.

try it in the sandbox
checkpoint auto 5checkpoint auto confirm

Goes withcheckpoint auto <minutes>

checkpoint auto <minutes>A change that undoes itself

Saves the running config now, and unless checkpoint auto confirm comes in time the switch puts it back. On my bench it rolled back about 5 s after the session that armed it ended, not only at the timer, so a change that cuts your SSH session comes straight back. Every change made meanwhile gets a WARNING asking for the confirm.

In every release here, 10.15 to 10.18.

Habit: Arm it before anything that touches how you reach the switch: the management address, an ACL on the management VRF, the login list. A management address change made over SSH always comes back under it; make that one from the console.
try it in the sandbox
checkpoint auto 5configure terminalvlan 60name CAMERASendcheckpoint auto confirm

Goes withcheckpoint auto confirmcheckpoint rollback <WORD>show checkpoint

VLANs4

vlan <1-4094>Create a VLAN

Creates the VLAN and enters its context. Name it there. A port cannot use a VLAN that does not exist.

In every release here, 10.15 to 10.18.

try it in the sandbox
vlan 10name STAFFvlan 30name VOICEvoice

Goes withshow vlanname <WORD>

name <WORD>Name the VLAN

Shows in show vlan and in LLDP. One word here; the box takes a longer line too.

In every release here, 10.15 to 10.18.

try it in the sandbox
vlan 20name PRINTERSvlan 99name AP-MGMT
voiceMark the voice VLAN

Tells LLDP-MED this VLAN carries voice. Phones that speak LLDP-MED learn it from the network policy the switch sends on ports that tag it: VLAN, priority 6, DSCP 46 with nothing else configured.

In every release here, 10.15 to 10.18.

Habit: Tag the voice VLAN on the phone port and leave the data VLAN native, so the PC behind the phone needs nothing.
try it in the sandbox
vlan 30voiceshow vlan voice

Goes withlldp med network-policyshow lldp neighbor-info

show vlanVLANs and their ports

Every VLAN, up or down, and which ports carry it. Down with no_member_port means no port is in it; no_member_forwarding means ports are in it but none is up. A role's VLAN is per client, MAC-based: the lab switch mapped each client into its role's VLAN on the port, and that VLAN's SVI came up with the first client.

In every release here, 10.15 to 10.18.

try it in the sandbox
show vlanshow vlan 30show vlan | include VOICE

Interfaces12

interface <IFNAME>Configure a port or a range

Enters one port (1/1/5) or a range (1/1/1-1/1/8). Member/slot/port: 1/1/5 is member 1, slot 1, port 5.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/5description printerinterface 1/1/1-1/1/8no shutdown
shutdownTurn a port off

Admin down. no shutdown turns it back on. A port BPDU guard disabled needs shutdown and then no shutdown, or the bpdu-guard timeout. A port nobody configured may be shut: show interface brief says Administratively down.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/11shutdowninterface 1/1/8no shutdown
description <LINE>Say what is on the port

Free text on the port, shown in show interface brief and LLDP.

In every release here, 10.15 to 10.18.

Habit: Describe uplinks and anything odd. The next person reads the brief, not your ticket.
try it in the sandbox
interface 1/1/7description desk 7, pair 3-6 open at 23 minterface lag 1description uplink to core
no routingMake it a switch port

A routed port takes an ip address; a switched port takes VLANs. no routing makes it switched. Some platforms (and the Switch Simulator) start every port routed, so check before you assign a VLAN.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/1no routingvlan access 10interface 1/1/12routingip address 203.0.113.129/25
vlan access <1-4094>One untagged VLAN

Makes the port an access port in that VLAN. The VLAN has to exist.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/5vlan access 20interface 1/1/1-1/1/4vlan access 10
vlan trunk allowed <VLIST>Tag VLANs on a port

Makes the port a trunk and adds VLANs to its allowed list. It adds: vlan trunk allowed 10 then vlan trunk allowed 20 leaves 10,20, as the lab switch did. no vlan trunk allowed 10 takes one out.

In every release here, 10.15 to 10.18.

Habit: A trunk prints vlan trunk native 1 even when you never set it. Set the native VLAN on purpose.
try it in the sandbox
interface 1/1/3vlan trunk native 10vlan trunk allowed 10,30interface lag 1vlan trunk allowed 10,20,30

Goes withvlan trunk native <1-4094>

vlan trunk native <1-4094>The untagged VLAN on a trunk

The VLAN whose frames cross the trunk untagged. On a phone port it is the PC's data VLAN; the phone tags voice.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/3vlan trunk native 10vlan trunk allowed 30interface 1/1/13vlan trunk native 99vlan trunk allowed 10,20,99
speed autoPort speed and auto-negotiationsandbox fakes

speed auto offers every speed the port can do; speed auto 1g 2.5g offers only those; speed 1000-full fixes it and turns auto-negotiation off. Since 10.09 the CLI hides speeds the hardware lacks.

In every release here, 10.15 to 10.18.

Where the sandbox pretends: The sandbox models the hardware. The Switch Simulator has virtual ports and hides speed entirely; the syntax is from HPE's fundamentals guide.
Habit: Leave Smart Rate ports on speed auto. Pinning an AP to 1G to fix a flap hides the real problem, usually the cable.
try it in the sandbox
interface 1/1/1speed autointerface 1/1/1speed auto 1g 2.5g 5ginterface 1/1/2no speed

Goes withshow interface <PORT> physicaldiag cable-diagnostic test <PORT>

show interface briefEvery port on one screen

Native VLAN, mode, enabled, up or down, why it is down, speed and description. Read the Reason column first, but don't trust an empty one: on 10.18 a port BPDU guard or loop protect shut shows down with no reason. show spanning-tree and show loop-protect say why.

In every release here, 10.15 to 10.18.

try it in the sandbox
show interface briefshow interface brief | include down
show interface <PORT>One port in detailchanged

State, why it is down (State information), VLANs and counters. Blocked by Port Access Security means the link is up but no client has authenticated.

In every release here, 10.15 to 10.18.

Changed: 10.15 says how long the link has been up; 10.16 says up; 10.17 added the Hardware port line.
try it in the sandbox
show interface 1/1/5show interface 1/1/5 | include State
show interface <PORT> physicalSpeed, PoE and statesandbox fakes

Link and admin state, the speed it runs at next to what it is configured to allow, flow control, PoE draw.

In every release here, 10.15 to 10.18.

Where the sandbox pretends: The PoE column is the sandbox's: the lab's virtual ports draw nothing.
try it in the sandbox
show interface 1/1/1 physicalshow interface physical
show mac-address-tableWho is behind which port

Learned MACs by VLAN and port. port-access-security means the MAC was learned through port-access. Filter by vlan or interface, or pipe it.

In every release here, 10.15 to 10.18.

try it in the sandbox
show mac-address-tableshow mac-address-table interface 1/1/6show mac-address-table | include 53:4c

LAGs and LACP6

LACP bringing up a two-member LAG, and the flags that prove it.diagram
LACP: both ends have to agree before a member carries traffic access-01 interface lag 1 lacp mode active members 1/1/13-14 2 core interface lag 1 lacp mode active members 1/1/49-50 LACPDUs both ways on each member: every 30 s, or 1 s with lacp rate fast 3 1. Hello: both ends active each sends system id, key and port A L F N C D 2. In sync: same aggregation each agrees the other belongs in lag 1 A L F N C D 3. Up: collecting, distributing only now does the member carry traffic A L F N C D 1 A active, P passive · L long timeout, S short · F aggregable · N in sync · C collecting · D distributing 1 Passive on both ends: nobody starts, nothing forms. One end active is enough; both active is the habit. 2 A new LAG on 10.18 is administratively down. Nothing forms until no shutdown on interface lag 1. 3 No lacp mode means a static LAG: no LACPDUs, so a member cabled to the wrong box still forwards. Proof: show lacp interfaces reads ALFNCD for actor and partner on every member; show lacp aggregates lists them.
  • Build the LAG first, then add members, then no shutdown the LAG. Settings live on the LAG.
  • lacp mode active on both ends. A static LAG only where the far end cannot speak LACP.
  • lacp rate fast on both ends notices a dead member in about three seconds instead of ninety.
  • Done means ASFNCD for actor and partner on every member in show lacp interfaces, or ALFNCD if you left the rate slow.
interface lag <1-256>Create a link aggregation

Creates lagN and enters it. On 10.18 a new LAG is administratively down: nothing forms until no shutdown.

In every release here, 10.15 to 10.18.

Habit: Configure the LAG first, then put ports in it. Settings live on the LAG, not the members.
try it in the sandbox
interface lag 1no shutdownno routingvlan trunk allowed 10,20lacp mode active

Goes withlag <1-256>lacp mode <active|passive>

lag <1-256>Put a port in a LAG

Makes the port a member. Its VLAN settings come from the LAG from then on.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/13-1/1/14lag 1
lacp mode <active|passive>Run LACP

active sends LACPDUs; passive only answers. Two passive ends never form a LAG. Without lacp mode the LAG is static: no LACP at all.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface lag 1lacp mode activeinterface lag 1lacp mode passive
lacp rate <fast|slow>LACPDUs every second

Asks the partner for a short timeout: a dead link is noticed in about three seconds instead of ninety. The S in PSFNCD is the partner asking for the same.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface lag 1lacp rate fastinterface lag 1no lacp rate
show lacp interfacesDid the partner agreechanged

Actor and partner state per member. ALFNCD is what you want: Active, Long timeout, aggregable, in sync, collecting, distributing. With lacp rate fast the L reads S, so ASFNCD is just as good. An empty partner line means nothing is talking back.

In every release here, 10.15 to 10.18.

Changed: 10.18 added IE (LACP fallback active) to the legend; 10.17 has no such line.
try it in the sandbox
show lacp interfacesshow lacp interfaces 1/1/13
show lacp aggregatesEach LAG, its members and mode

Members, heartbeat rate, hash and mode per LAG.

In every release here, 10.15 to 10.18.

try it in the sandbox
show lacp aggregatesshow interface lag 1

Spanning tree5

spanning-treeTurn on spanning tree

MSTP by default. Without it an accidental loop takes the VLAN down. mode rpvst switches to Rapid PVST+.

In every release here, 10.15 to 10.18.

Habit: Decide who is root and set its priority low. Leave it to chance and the lowest bridge MAC wins, which is usually the oldest switch in the closet.
try it in the sandbox
spanning-treespanning-treespanning-tree priority 4
spanning-tree port-type admin-edgeAn edge port

A port where only end devices live: it goes straight to forwarding instead of waiting out discarding and learning.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/1-1/1/12spanning-tree port-type admin-edge

Goes withspanning-tree bpdu-guard

spanning-tree bpdu-guardShut the port if a switch appears

A BPDU arriving on the port disables it. show spanning-tree calls it Bpdu-Error; on 10.18 show interface brief leaves the Reason column empty, so don't look there. Pair it with admin-edge on every access port. Once the switch is gone, shutdown then no shutdown brings the port back, or spanning-tree bpdu-guard timeout does it on its own.

In every release here, 10.15 to 10.18.

Habit: Access ports: admin-edge plus bpdu-guard. Uplinks: neither.
try it in the sandbox
interface 1/1/1-1/1/12spanning-tree bpdu-guard
loop-protectCatch loops spanning tree cannot see

Sends probes and error-disables the port if one comes back, which catches a loop through an unmanaged switch that drops BPDUs.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/1-1/1/12loop-protect
show spanning-treeRoot, role and state per port

Who is root, which port leads there, and each port's role and state. Under MSTP it prints only instance 0, the CIST; show spanning-tree mst <n> gives the roles in another instance.

In every release here, 10.15 to 10.18.

try it in the sandbox
show spanning-treeshow spanning-tree | include Root

LLDP3

LLDP-MED on a phone port: the phone speaks first, the switch answers with the voice VLAN.diagram
LLDP-MED: the phone speaks first, then the switch hands it the voice VLAN IP phone MED endpoint, PC behind it Switch 1/1/3 network connectivity device LLDPDU with LLDP-MED TLVs capabilities, class III, inventory, PoE 2 LLDPDU with a network policy 1 voice: VLAN 30, tagged, prio 6, DSCP 46 Phone tags voice 30, DHCPs there voice frames, tagged 30 PC frames, untagged: native VLAN 10 ON THE SWITCH vlan 30 name VOICE voice interface 1/1/3 vlan trunk native 10 vlan trunk allowed 10,30 lldp med network-policy is on by default on every port WHAT PROVES IT show lldp neighbor-info 1/1/3 show vlan voice show mac-address-table 1 The policy goes out only after the switch hears the phone's MED TLVs. 10.18 added lldp med force-send. 2 On a port-access port, LLDP from a device that has not authenticated is dropped. allow-lldp-bpdu lets it in. No MED from the phone, no network policy back: check the phone's LLDP setting before the switch.
  • Make the voice VLAN a real VLAN with voice set, tag it on the phone port, and leave the data VLAN native.
  • No network policy on the phone? Check the phone sends LLDP-MED before touching the switch.
  • On a NAC port add allow-lldp-bpdu, or LLDP from the phone is dropped before anything reads it.
lldp med network-policySend voice VLAN policy to phoneschanged

On by default. On a port that tags a voice VLAN the switch sends LLDP-MED network policy, but only after it has heard the endpoint's own LLDP-MED TLVs. no lldp med network-policy stops it.

In every release here, 10.15 to 10.18.

Changed: New in 10.18: lldp med force-send. By default MED TLVs go out only after the endpoint's own arrive (HPE's guide and the lab agree); the sandbox does not model force-send.
try it in the sandbox
interface 1/1/3lldp med network-policyinterface 1/1/3no lldp med network-policy

Goes withvoice

lldp <transmit|receive>LLDP on a port

LLDP sends and listens on every port by default. no lldp transmit and no lldp receive turn each half off.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/9no lldp transmitinterface 1/1/9lldp transmitlldp receive
show lldp neighbor-infoWho is plugged in where

The LLDP table: port, chassis, remote port, name. Give a port for the full record: capabilities, power it asks for, LLDP-MED class and network policy.

In every release here, 10.15 to 10.18.

Habit: On a port-access port the switch drops LLDP from a client that has not authenticated, unless the port has allow-lldp-bpdu. An empty neighbor on a phone port is often that.
try it in the sandbox
show lldp neighbor-infoshow lldp neighbor-info 1/1/3show lldp neighbor-info | include CAM

RADIUS7

802.1X against ClearPass, from link up to a change of authorization.diagram
802.1X against ClearPass: who says what, in order Laptop supplicant Switch 1/1/1 authenticator, relays only ClearPass RADIUS server, decides EAP-Request, Identity EAPOL on the wire; link came up EAP-Response, Identity the outer identity only RADIUS Access-Request 1 EAP inside, UDP 1812, shared secret PEAP or EAP-TLS: Access-Challenge and Access-Request, several round trips the TLS tunnel runs from the laptop to ClearPass; the switch never sees inside it RADIUS Access-Accept Aruba-User-Role = EMPLOYEE EAP-Success port opens in the role's VLAN CoA-Request, UDP 3799 2 new role, or end the session CoA-ACK 1 A wrong shared secret looks like a dead server: ClearPass drops the request and the switch counts a timeout. 2 On 10.18 a CoA needs radius dyn-authorization client with that server's address, or it is dropped as invalid.
  • Two RADIUS servers in the group before go-live. One server is a single point of failure for every port.
  • A critical role on every port-access port, so a ClearPass outage ends somewhere you chose.
  • radius dyn-authorization client for each ClearPass node, or the switch drops every CoA and Disconnect and the only trace is the Invalid Client Addresses counter in show radius dyn-authorization.
  • Auth failing? Look in Access Tracker first. No entry at all means ClearPass never processed it: check the address, the VRF and the shared secret. A wrong secret reads Server-Timeout on the switch, same as a dead server.
radius-server host <A.B.C.D> key plaintext <WORD>Point the switch at ClearPass

Defines a RADIUS server and the shared secret. The running config shows the key as ciphertext. A wrong key looks like a dead server: the server drops the request and the switch times out.

In every release here, 10.15 to 10.18.

Habit: Use a group (aaa group server radius) even with one server, so adding the second is one line.
try it in the sandbox
radius-server host 192.0.2.10 key plaintext cppm-lab-keyradius-server host 192.0.2.11 key plaintext cppm-lab-key vrf mgmt

Goes withaaa group server radius <WORD>show radius-server

aaa group server radius <WORD>A group of RADIUS serverschanged

Port-access points at a group, not a server. Servers are tried in priority order; entered without a priority they number themselves in the order you add them.

In every release here, 10.15 to 10.18.

Changed: 10.16 added server <ip> priority <n>; 10.15 has no priority in a group.
Changed: 10.18 added load-balancing-mode (round-robin or priority-based) in the group.
try it in the sandbox
aaa group server radius CLEARPASSserver 192.0.2.10server 192.0.2.11aaa group server radius CLEARPASSserver 192.0.2.11 priority 1

Goes withradius-server host <A.B.C.D> key plaintext <WORD>

show radius-serverServers and whether they answer

Each server with its ports and VRF. A star in front means the switch has marked it unreachable.

In every release here, 10.15 to 10.18.

try it in the sandbox
show radius-servershow radius-server detail

Goes withshow radius-server statistics authentication

show radius-server statistics authenticationRequests, accepts, rejects, timeouts

Counters per server. Timeouts climbing with no accepts usually means the shared secret. All zeros while clients hit Server-Timeout means the requests never left: on my bench a server the switch couldn't ARP for never moved a counter. show radius-server statistics on its own is incomplete on 10.18.

In every release here, 10.15 to 10.18.

try it in the sandbox
show radius-server statistics authenticationshow radius-server statistics authentication | include Timeouts
radius dyn-authorization enableListen for CoA

Lets ClearPass change or end a session after the fact (RFC 5176). On 10.18 that is two pieces: enable it, and name each server allowed to send with radius dyn-authorization client. Without the client line requests are dropped and counted as invalid client addresses.

In every release here, 10.15 to 10.18.

Habit: Pick the AOS-CX profiles in ClearPass: the AOS-CX Disconnect profile ends a session and the AOS-CX Bounce Switch Port profile is the CoA. The default ArubaOS Wireless Terminate Session profile carries only Calling-Station-Id, and the switch NAKs it with Missing-Attribute. Tested against ClearPass 6.11.1 on 2026-09-29.
try it in the sandbox
radius dyn-authorization enableradius dyn-authorization client 192.0.2.10 secret-key plaintext cppm-lab-key

Goes withshow radius dyn-authorization

show radius dyn-authorizationCoA counters and clients

Whether dynamic authorization is on, the port (3799), the invalid-address counters, and per client the CoA and disconnect requests, ACKs and NAKs.

In every release here, 10.15 to 10.18.

try it in the sandbox
show radius dyn-authorization
show aaa server-groupsEvery server group, RADIUS and TACACS

Each group with its servers, port, VRF and priority, TACACS+ first and RADIUS after it. The switch's own groups, radius and tacacs, sit at the bottom of each half and hold every server you configured. show aaa server-groups tacacs prints only the TACACS half.

In every release here, 10.15 to 10.18.

try it in the sandbox
show aaa server-groupsshow aaa server-groups tacacs

Goes withaaa group server tacacs <name>

Management access10

tacacs-server host <A.B.C.D> key plaintext <key>A TACACS+ server for admin logins

One line per ClearPass node, with the shared secret the network device has in ClearPass. The servers go into a group, and the group goes into a channel's login list. tacacs-server lines print above radius-server lines in the running config.

In every release here, 10.15 to 10.18.

try it in the sandbox
tacacs-server host 192.0.2.10 key plaintext cppm-tac-keytacacs-server host 192.0.2.11 key plaintext cppm-tac-key vrf mgmt

Goes withaaa group server tacacs <name>aaa authentication login <default|ssh|console|https-server> <methods>

aaa group server tacacs <name>The TACACS servers a login list asks

The servers in the order they are asked. The switch keeps a group called tacacs of its own with every tacacs-server in it, the way it keeps one called radius; name yours something else.

In every release here, 10.15 to 10.18.

try it in the sandbox
tacacs-server host 192.0.2.10 key plaintext cppm-tac-keyaaa group server tacacs CPPM-TACserver 192.0.2.10

Goes withtacacs-server host <A.B.C.D> key plaintext <key>show aaa server-groups

aaa authentication login <default|ssh|console|https-server> <methods>How admins log in, per channel

The methods a channel tries, in order: group <name> for TACACS, then local. Local is used when every server in the groups times out. When ClearPass answers no, that is the answer, unless allow-fail-through is on. Leave console on local so there is always a way in.

In every release here, 10.15 to 10.18.

Habit: The ClearPass profile needs a Shell attribute, priv-lvl (15 or 1) or Aruba-Admin-Role. A Privilege Level in the profile header alone gave Permission denied on my bench while Access Tracker logged ACCEPT.
try it in the sandbox
tacacs-server host 192.0.2.10 key plaintext cppm-tac-keyaaa group server tacacs CPPM-TACserver 192.0.2.10exitaaa authentication login ssh group CPPM-TAC localaaa authentication login console local

Goes withaaa authentication allow-fail-throughshow aaa authentication

aaa authentication allow-fail-throughLet a reject try the next method

A reject from ClearPass falls through to the next method, local. On my bench that is what let the switch's own admin in over SSH while ClearPass was up. It also means the local password works whenever ClearPass says no, so I leave it off and use the console for the local account.

In every release here, 10.15 to 10.18.

try it in the sandbox
aaa authentication allow-fail-throughno aaa authentication allow-fail-through

Goes withaaa authentication login <default|ssh|console|https-server> <methods>

aaa authorization commands <default|ssh|console> <methods>Ask ClearPass before every command

Every command a session runs is asked of ClearPass first, and the switch asks y/n before it turns this on. On my bench a user ClearPass didn't know, the local admin included, then couldn't run anything, show clock included, and it came back out from a TACACS administrator's session. none on the end lets commands run when no server answers at all.

In every release here, 10.15 to 10.18.

Habit: Turn it on last, from a TACACS session, once ClearPass has command rules for everyone who logs in. Keep a second session open while you do.
try it in the sandbox
tacacs-server host 192.0.2.10 key plaintext cppm-tac-keyaaa group server tacacs CPPM-TACserver 192.0.2.10exitaaa authorization commands ssh group CPPM-TAC noneno aaa authorization commands ssh

Goes withshow aaa authorizationaaa authentication login <default|ssh|console|https-server> <methods>

aaa accounting all-mgmt <default|ssh|console|https-server> start-stop <methods>A record of who logged in and what they ran

Sends logins and commands to the TACACS group, and with local on the end keeps a copy on the switch too, so the record survives ClearPass being down (CIS 1.5.4.1).

In every release here, 10.15 to 10.18.

try it in the sandbox
tacacs-server host 192.0.2.10 key plaintext cppm-tac-keyaaa group server tacacs CPPM-TACserver 192.0.2.10exitaaa accounting all-mgmt ssh start-stop group CPPM-TAC local

Goes withaaa authentication login <default|ssh|console|https-server> <methods>

show aaa authenticationThe login lists, per channel

Fail-through and each channel's methods in priority order, 0 asked first. The default channel stays local unless you change it, and a channel with nothing of its own uses default.

In every release here, 10.15 to 10.18.

try it in the sandbox
show aaa authentication

Goes withaaa authentication login <default|ssh|console|https-server> <methods>

show aaa authorizationWho decides what a session may run

Command authorization per channel. local on its own is the switch's own rules for the group a user landed in; a TACACS group means ClearPass is asked for every command.

In every release here, 10.15 to 10.18.

try it in the sandbox
show aaa authorization

Goes withaaa authorization commands <default|ssh|console> <methods>

show user informationWho this session is

The user, how they authenticated, the group the switch put them in and the privilege level. On my bench TACACS users landed in administrators at 15, operators at 1, auditors at 19, and a priv-lvl the switch has no group for came out as a group of that number with privilege N/A that could run nothing.

In every release here, 10.15 to 10.18.

try it in the sandbox
show user information

Goes withaaa authentication login <default|ssh|console|https-server> <methods>

interface mgmtThe out-of-band management portsandbox fakes

ip dhcp by default, or ip static with default-gateway. Change the address from your own SSH session to it and the session goes with it, which is what checkpoint auto is for and why I make this change from the console.

In every release here, 10.15 to 10.18.

Where the sandbox pretends: The sandbox models one SSH session, yours, to the DHCP address; sim console is the console port.
try it in the sandbox
sim consoleconfigure terminalinterface mgmtip static 192.0.2.51/24default-gateway 192.0.2.1

Goes withcheckpoint auto <minutes>

Port access17

Multi-domain authentication: a phone and the PC behind it, each with its own role.diagram
One port, a phone and a PC: multi-domain authentication PC plugged into the phone IP phone no supplicant Switch 1/1/6 auth-mode multi-domain 1 ClearPass names the roles VOICE DOMAIN Phone MAC auth, or an LLDP device profile 3 Accept, role VOICE device-traffic-class voice 2 The voice device VLAN 30 tagged, 10 native DATA DOMAIN PC 802.1X, straight through the phone Accept, role EMPLOYEE vlan access 10 The data device VLAN 10 untagged show port-access clients lists both on 1/1/6. show vlan counts 1/1/6 in 10 and in 30. 1 A port takes one client by default. The PC is refused until the port is multi-domain or client-limit 2. 2 The voice device is whoever holds a voice-class role. Without it the phone is one more data client. 3 LLDP from a phone that has not authenticated is dropped, so a device profile never sees it without allow-lldp-bpdu. A client that fails authentication does not use up a place; only the ones that got in count.
  • Multi-domain is one voice device plus one data device. More PCs behind a phone: raise client-limit multi-domain.
  • The phone's role carries device-traffic-class voice, a native data VLAN and the voice VLAN tagged.
  • Test the failure path before users do: take ClearPass away and read what the phone and the PC get.
aaa authentication port-access dot1x authenticator802.1X, switch-wide

Global switch for 802.1X plus the RADIUS group it uses. It needs enable here and on each port: either one off and nothing authenticates.

In every release here, 10.15 to 10.18.

try it in the sandbox
aaa authentication port-access dot1x authenticatorradius server-group CLEARPASSenableinterface 1/1/1-1/1/8aaa authentication port-access dot1x authenticatorenable

Goes withaaa authentication port-access mac-authshow port-access clients

aaa authentication port-access mac-authMAC authentication, switch-wide

The same pattern as 802.1X: a global block with the group and enable, and enable per port. The switch sends the MAC as username and password, 00005e005305 style by default.

In every release here, 10.15 to 10.18.

try it in the sandbox
aaa authentication port-access mac-authradius server-group CLEARPASSenableinterface 1/1/5aaa authentication port-access mac-authenable
aaa authentication port-access client-limit <1-256>How many clients a port takeschanged

One by default, up to 256 on a 6200 or 6300. A PC behind a phone needs two, or a multi-domain port.

In every release here, 10.15 to 10.18.

Changed: 10.18 added client-limit device-mode <n> on the port and in a role.
try it in the sandbox
interface 1/1/3aaa authentication port-access client-limit 2
aaa authentication port-access auth-precedence <order>Which method goes first

dot1x then mac-auth is the default. A device with no supplicant waits for 802.1X to give up before MAC auth starts, and on 10.18 defaults that took 165 seconds on my bench. initial-auth-response-timeout 10 under the port's dot1x authenticator cut it to about 10. mac-auth first gets a printer in within seconds, but a MAC-auth accept ends it there and the laptop never tries 802.1X, so only put it first where a MAC-auth role is fine for everything on the port.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/5aaa authentication port-access auth-precedence mac-auth dot1xinterface 1/1/1aaa authentication port-access auth-precedence dot1x mac-auth
aaa authentication port-access critical-role <WORD>A role when RADIUS is down

If every RADIUS server times out, clients get this role instead of nothing. reject-role is the same for a reject.

In every release here, 10.15 to 10.18.

Habit: Decide what a building does when ClearPass is unreachable before it happens. Critical role is that decision written down.
try it in the sandbox
interface 1/1/1aaa authentication port-access critical-role QUARANTINEinterface 1/1/1aaa authentication port-access reject-role GUEST
aaa authentication port-access auth-mode multi-domainA phone and a PC on one port

One voice device and one data device (client-limit multi-domain raises the data count). The voice device is whoever holds a role with device-traffic-class voice.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/6aaa authentication port-access auth-mode multi-domaininterface 1/1/6aaa authentication port-access client-limit multi-domain 2

Goes withdevice-traffic-class voiceaaa authentication port-access allow-lldp-bpdu

aaa authentication port-access allow-lldp-bpduHear LLDP before authenticationchanged

A port-access port drops LLDP from a client that has not authenticated. A phone that never authenticates is then invisible to LLDP, its device profile never matches, and it gets no voice VLAN. This lets LLDP through. Found the hard way on the lab switch on 2026-09-28.

In every release here, 10.15 to 10.18.

Changed: 10.17 added aaa authentication port-access lldp-loop-guard enable beside it.
try it in the sandbox
interface 1/1/6aaa authentication port-access allow-lldp-bpdu

Goes withdebug <portaccess|radius|lldp> <what>

port-access role <WORD>A local user rolechanged

What a client gets once it is in: an access VLAN, or a native VLAN and a tagged list, and more on 10.18. RADIUS names the role; the switch must have it or the client fails authorization.

In every release here, 10.15 to 10.18.

Changed: 10.18 added speed, voice-vlan, stp-bpdu-guard, rate-limits, toggle-link and client-limit device-mode to a role.
try it in the sandbox
port-access role EMPLOYEEvlan access 10port-access role AP-TRUNKvlan trunk native 99vlan trunk allowed 10,20,99

Goes withshow port-access role

device-traffic-class voiceThis role is the voice device

On a multi-domain port the client holding this role is the voice device. Pair it with a native data VLAN and the voice VLAN tagged.

In every release here, 10.15 to 10.18.

try it in the sandbox
port-access role VOICEdevice-traffic-class voicevlan trunk native 10vlan trunk allowed 30
show port-access clientsWho is on and how

One line per client: port, name, role, VLAN, and four flags: how it got in (1x, ma, ps, dp), mode (c, d, m), device type (d, v) and status (s, f, p, and d for a downloadable role that failed to download). --|c|-|f is a failure. Add detail for the history, interface, mac or role to filter.

In every release here, 10.15 to 10.18.

Habit: Read Auth History in the detail view bottom to top: it is the order things happened.
try it in the sandbox
show port-access clientsshow port-access clients interface 1/1/6 detailshow port-access clients onboarding-method device-profile
show aaa authentication port-access interface all client-statusClient status, compact

The authentication and authorization blocks for every client, without VLAN and MACsec detail.

In every release here, 10.15 to 10.18.

try it in the sandbox
show aaa authentication port-access interface all client-statusshow aaa authentication port-access interface 1/1/1 client-status
show port-access roleThe roles and what they carry

Each local role: VLANs, device type, gateway zone. show port-access role name X for one.

In every release here, 10.15 to 10.18.

try it in the sandbox
show port-access roleshow port-access role name VOICE
port-access reauthenticate interface <IFNAME>Make clients authenticate again

10.18 has no clear port-access clients. To start clients over: reauthenticate a port, or log one off by MAC, port or role.

In every release here, 10.15 to 10.18.

try it in the sandbox
port-access reauthenticate interface 1/1/1port-access log-off client mac 00:00:5e:00:53:01port-access log-off client role GUEST
port-access fallback-role <WORD>A role for whatever is on the port

The role a client gets when nothing else gives it one. With no authentication on the port that is everything plugged in: with a gateway zone in the role, that is port-based tunneling.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/4port-access fallback-role PBT-IOT
initial-auth-response-timeout <seconds>How long a port waits for a first EAP answersandbox fakes

Under the port's dot1x authenticator. A device with no 802.1X client (a printer, a camera, a phone without a certificate) never answers, so this is how long the port waits before it moves on to MAC auth. On my bench the 10.18 defaults made a silent printer wait about 165 seconds, and 10 here got it on in about 10. A client that answers and then stalls ignores it.

In every release here, 10.15 to 10.18.

Where the sandbox pretends: The wait itself plays out only in labs with the sandbox clock (sim wait); elsewhere the line is just taken.
Habit: Set it wherever something without a supplicant can land. It costs a real laptop nothing: a supplicant answers inside a second.
try it in the sandbox
interface 1/1/3aaa authentication port-access dot1x authenticatorinitial-auth-response-timeout 10

Goes witheapol-timeout <seconds>aaa authentication port-access auth-precedence <order>

eapol-timeout <seconds>How long to wait for each EAP replysandbox fakes

Under the port's dot1x authenticator, with max-eapol-requests beside it. This is the one for a supplicant that starts 802.1X and then goes quiet halfway. On my bench a stalled PEAP client held the port about 164 s on the defaults, 122 s with eapol-timeout 10, and 42 s with eapol-timeout 10 plus max-eapol-requests 1.

In every release here, 10.15 to 10.18.

Where the sandbox pretends: The sandbox's times are those three measurements and a line fitted through them; other combinations are the fit, not a measurement.
try it in the sandbox
interface 1/1/4aaa authentication port-access dot1x authenticatoreapol-timeout 10max-eapol-requests 1

Goes withmax-eapol-requests <1-10>initial-auth-response-timeout <seconds>

max-eapol-requests <1-10>How many times to ask before giving up

How many EAP requests the port sends to a client that has gone quiet before it calls 802.1X failed and moves on. Down to 1 beside eapol-timeout 10 took a stalled client from about 164 s to about 42 s on the lab switch.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface 1/1/4aaa authentication port-access dot1x authenticatormax-eapol-requests 1

Goes witheapol-timeout <seconds>

Tunneling4

gateway-zone zone <WORD> gateway-role <WORD>Tunnel this role's traffic

Clients in this role are tunneled to the gateway cluster of the zone, where the gateway role applies. That is user-based tunneling when RADIUS assigns the role. Put the same kind of role on a port as its fallback role and everything on the port tunnels; the sandbox models that, and I haven't run it against a real gateway.

In every release here, 10.15 to 10.18.

try it in the sandbox
port-access role TUNNEL-EMPLOYEEgateway-zone zone CAMPUS gateway-role authenticatedport-access role PBT-IOTgateway-zone zone CAMPUS gateway-role iot

Goes withubt zone <WORD> vrf <WORD>port-access fallback-role <WORD>

ubt-client-vlan <1-4094>The VLAN tunneled clients use

In local-vlan mode (what show ubt reports) this is the reserved VLAN all tunneled client traffic uses to reach the gateway. Create it first and use it for nothing else, as HPE's guide says.

In every release here, 10.15 to 10.18.

try it in the sandbox
vlan 666name UBT-TRANSPORTexitubt-client-vlan 666

Goes withubt zone <WORD> vrf <WORD>

ubt zone <WORD> vrf <WORD>Where the tunnels gochangedsandbox fakes

A zone names the gateway cluster: primary and backup controller addresses, reached in a VRF, and enable. show ubt says whether it is up.

In every release here, 10.15 to 10.18.

Changed: 10.16 took IPv6 controller addresses.
Where the sandbox pretends: The gateway is a bench device in the sandbox; a zone is up when the switch can reach it.
try it in the sandbox
ubt zone CAMPUS vrf defaultprimary-controller ip 192.0.2.50backup-controller ip 192.0.2.51enable
show ubtZone state

Each zone's controllers, VLAN, admin and operational state. brief is one line per zone; users all lists who is tunneled.

In every release here, 10.15 to 10.18.

try it in the sandbox
show ubtshow ubt briefshow ubt users all

Device profiles3

port-access lldp-group <WORD>Recognize a device by its LLDPchanged

Ordered match and ignore rules; the first rule a neighbor hits decides. sys-desc matches if the description contains the word (proved on the lab switch), vendor-oui plus type matches a TLV.

In every release here, 10.15 to 10.18.

Changed: 10.18 added match chassis-id. sys-desc, sysname and vendor-oui are there from 10.15.
try it in the sandbox
port-access lldp-group APSmatch sys-desc AP-515port-access lldp-group LLDP-MED-ENDPOINTSseq 10 match vendor-oui 0012bb type 1port-access lldp-group COREmatch sysname core1

Goes withport-access device-profile <WORD>

port-access device-profile <WORD>Give a recognized device a role

Ties an LLDP group to a role, then enable. Works on a port with no port-access at all; on a port-access port it applies when 802.1X and MAC auth do not get the device in. The client shows as dp.

In every release here, 10.15 to 10.18.

try it in the sandbox
port-access device-profile APSassociate lldp-group APSassociate role AP-TRUNKenable

Goes withshow port-access device-profile

show port-access device-profileProfiles and who matched

The profiles and their state. interface all lists each matched neighbor with its profile, group, role and whether it applied.

In every release here, 10.15 to 10.18.

try it in the sandbox
show port-access device-profileshow port-access device-profile interface all

Routing4

interface vlan <1-4094>An address on a VLAN

A switch virtual interface: the switch's own address in that VLAN, up only when a port in the VLAN is up.

In every release here, 10.15 to 10.18.

try it in the sandbox
interface vlan 10ip address 192.0.2.1/24
ip route <A.B.C.D/M> <A.B.C.D>A static route

Prefix and next hop. The default route is 0.0.0.0/0.

In every release here, 10.15 to 10.18.

try it in the sandbox
ip route 0.0.0.0/0 203.0.113.1ip route 198.51.100.0/24 203.0.113.1
router ospf <1-63>Start OSPF

The process, its router-id and areas. Interfaces join with ip ospf <process> area <area>. Both ends of a link must agree on the area.

In every release here, 10.15 to 10.18.

try it in the sandbox
router ospf 1router-id 203.0.113.2area 0.0.0.0exitinterface vlan 100ip ospf 1 area 0.0.0.0
show ip routeThe routing table

Connected (C), local (L), static (S) and OSPF (O) routes with next hop and interface.

In every release here, 10.15 to 10.18.

try it in the sandbox
show ip routeshow ip ospf neighborsping 198.51.100.10

Stacking2

vsf member <1-8>Stack another switchsandbox fakes

Virtual Switching Framework: pre-provision a member by part number and name the ports that form its stack links. A member's links are its own ports, 2/1/x for member 2. Member 1 is the switch you are on. vsf secondary-member picks the standby; without it there is none.

Not in the Switch Simulator's command set for 10.15 to 10.18. Syntax from HPE's guides.

Where the sandbox pretends: The Switch Simulator has no VSF, so nothing here was checked against a capture. The syntax is from HPE's VSF guide; the sandbox never cables a second switch.
try it in the sandbox
vsf member 2type jl725alink 1 2/1/27exitvsf secondary-member 2vsf member 1link 1 1/1/16

Goes withshow vsf

show vsfThe stack and its memberssandbox fakes

Each member with its role (conductor, standby, member), status and part number. show vsf link lists the stack links and whether they are up.

Not in the Switch Simulator's command set for 10.15 to 10.18. Syntax from HPE's guides.

Where the sandbox pretends: The layout is the sandbox's own: the Switch Simulator has no VSF to capture. A provisioned member shows Not Present because no second switch is ever cabled.
try it in the sandbox
show vsfshow vsf link

Diagnostics3

debug <portaccess|radius|lldp> <what>Log what port-access does

Writes port-access events to the debug buffer, the default destination. debug radius all and debug lldp event add those modules. show debug lists what is on; no debug all turns it off.

In every release here, 10.15 to 10.18.

Habit: Clear the buffer, turn debug on, reproduce once, read, turn it off. A busy switch fills the buffer in seconds.
try it in the sandbox
debug portaccess alldebug lldp eventshow debug buffer module portaccessshow debug buffer module lldp | include 1/1/6

Goes withshow debug buffer

show debug bufferRead the debug logsandbox fakes

Everything logged, oldest first. module portaccess, radius or lldp narrows it; a pipe narrows it further.

In every release here, 10.15 to 10.18.

Where the sandbox pretends: The sandbox logs the lines that tell a client's story, in the lab's format; the box logs far more.
try it in the sandbox
show debug buffershow debug buffer module portaccess | include DEVICEPROFILEclear debug buffer
diag cable-diagnostic test <PORT>Test the cable from the switchchangedsandbox fakes

A TDR test on a copper port: per pair good, open or short, the cable length or the distance to the fault. It drops the link while it runs and asks first. show prints the result; clear forgets it.

In every release here, 10.15 to 10.18.

Changed: Introduced in 10.11 (HPE's diagnostics guide).
Where the sandbox pretends: Faked hardware. The Switch Simulator has no PHY and refuses the command; the sandbox builds the result from the lab's description of the cable, in the guide's layout.
try it in the sandbox
diag cable-diagnostic test 1/1/7diag cable-diagnostic show 1/1/7diag cable-diagnostic clear 1/1/7

REST1

sim rest post /rest/v10.18/loginThe same switch through RESTsandbox fakessandbox only

sim rest is a laptop calling the REST API: log in, then GET, POST, PATCH or DELETE under /rest/v10.18, and POST /rest/v10.18/cli for AnyCLI, which runs only the exact commands GET /rest/v10.18/cli/commands lists. Each answer names its CLI twin.

A sandbox command, not a switch command.

Where the sandbox pretends: A sandbox command, not a switch command: the shapes are the lab switch's REST answers from 2026-09-28.
try it in the sandbox
sim rest post /rest/v10.18/loginsim rest get /rest/v10.18/system/vlans?depth=2&attributes=id,namesim rest post /rest/v10.18/cli {"cmd":"show vlan"}

Sandbox5

sim connect <DEV>Plug a device insandbox only

The sandbox talking, not the switch. sim connect and sim disconnect plug bench devices in and out, sim coa has the fake ClearPass send a change of authorization, sim release picks the AOS-CX release, sim status shows what is plugged in.

A sandbox command, not a switch command.

try it in the sandbox
sim statussim connect phonesim release 10.16
sim wait <seconds>Let time passsandbox only

The sandbox clock. Lets that many seconds pass in labs where timers matter, then says who got on.

A sandbox command, not a switch command.

try it in the sandbox
sim wait 30
sim ssh <user>Log this terminal in as someone elsesandbox only

Logs this terminal in again over SSH as that user. The fake ClearPass answers TACACS for the lab's users, and its profiles are in the ClearPass panel. sim ssh on its own says who the terminal is now.

A sandbox command, not a switch command.

try it in the sandbox
sim ssh
sim consoleWalk over to the console portsandbox only

Puts this terminal on the console as the local admin. The console uses its own login list, local unless you changed it, and doesn't care what the management port's address is.

A sandbox command, not a switch command.

try it in the sandbox
sim console
sim cppmAccess Tracker in the terminalsandbox only

The fake ClearPass's Access Tracker as a table, newest first. The ClearPass panel shows the same rows with their detail and Change Status. No row at all means the request never reached ClearPass.

A sandbox command, not a switch command.

try it in the sandbox
sim cppm