Documentation menu
Multi-agent mesh
Measuring the paths between your own sites, in both directions, as one test.
A mesh test measures path health between two and five of your own agents, in every direction, as a single test.
Why measure between your own sites
Because the paths between your offices are the ones you are accountable for, and they are asymmetric.
Traffic from Dublin to Frankfurt does not necessarily return by the route it took. When a carrier reroutes one direction, a measurement taken from one end sees a change and a measurement from the other end sees nothing, and the two teams spend an afternoon disagreeing. A mesh measures every direction, so the asymmetry is visible instead of being the thing that confuses everyone.
It is also the evidence that settles a conversation with a carrier. “Our monitoring says it is slow” is an opinion. Continuous per-hop measurement in both directions, with the route fingerprinted on every pass, is not.
Creating one
Multi-agent is a standalone check type: it cannot share a test with anything else, because its arrangement is agents rather than a target.
Select Multi-agent, then choose between two and five agents as members. Each member can carry an address override, which you need when the address an agent should be reached on differs from the one it reports, behind NAT, or on a management network that is not the path you want measured.
The address family must be explicit. Mesh expands into path-health legs, and those require a family for the same reason a standalone Path Health check does: a dual-stack pair measured on whichever family happened to win tells you nothing reliable about either.
The upper bound of five is a deliberate one. A mesh measures every direction between every pair, so the number of legs grows quadratically, and the difference between useful coverage and a self-inflicted probe volume problem arrives quickly.
3 agents, every pair, both directions = 6 legs:
A → B A → C B → C
A ← B A ← C B ← C
4 agents = 12 legs. 5 agents = 20 legs.Members are fixed after creation
You can rename the test and edit its labels afterwards, but the member list is set at creation and shown read-only in the Manage tab.
This is because the members define what the test is. Silently swapping one site for another would leave a single continuous history describing two different things, and every trend line crossing that boundary would be a lie. When the sites change, make a new test.
Reading the results
Mesh results include a leg selector, since there is one path per direction per pair. Choose a leg and you get the ordinary path view for it: per-hop loss, latency and jitter, route fingerprinting, and the timeline.
The comparison worth making is a leg against its reverse. Two directions between the same pair of sites that disagree about where loss appears is the signature of a routing asymmetry, and it is very hard to see any other way.