Active Probing and Its Limits

In one line: probing is the first design anyone reaches for, it is genuinely useful, and understanding exactly where it stops working is what motivates the real solution.

The initial design

Each prober periodically requests the service and records whether it succeeded. If European probers fail while others succeed, you have located a regional connectivity problem — from outside, which is exactly the vantage point the previous lesson showed was necessary.

This is active probing: you generate the traffic and observe the result.

Where it runs out

External probers face two significant limitations.

Incomplete coverage

The internet consists of over 100,000 autonomous systems. Placing probes in every one is cost-prohibitive and logistically difficult due to varying regulations and maintenance requirements.

Autonomous systems on the internet   > 100,000
Realistic prober deployments         tens to hundreds of vantage points
Coverage                             a tiny fraction of the paths users take

The coverage gap is not a rounding error

Probing moves the observer closer to the user without ever reaching them — which is why it complements the agent rather than replacing it.

With a few hundred vantage points against more than 100,000 autonomous systems, your probers observe a vanishingly small sample of the network paths your users actually use.

And the failures that matter most are precisely the ones concentrated in networks you have no probe in — a single regional ISP misrouting traffic affects every one of its customers and none of your probers. The previous lesson's regional outage is exactly this case.

Probing gives you presence, not coverage. It answers "is the service reachable from where I happen to be watching?", which is a weaker question than "can my users reach it?"

Synthetic traffic

Probes generate synthetic traffic, which may not accurately represent real user behavior or complex interaction patterns.

	A prober's request	A real user's session
Path	One endpoint, fixed	Many endpoints, varied
Client	A controlled agent	Every browser, OS, and device version
Network	A data center connection	Mobile, congested Wi-Fi, captive portals
Behavior	A simple, repeated check	Complex interaction patterns

A probe confirming that a health endpoint responds tells you little about whether a user on a mobile network behind a corporate proxy can complete a checkout.

Key takeaway

Active probing puts you outside the failure domain, which is the right instinct. Its ceiling is coverage — a few hundred vantage points cannot represent 100,000 autonomous systems, and synthetic traffic is not real traffic.

Interview signal by level

Level	What a strong answer sounds like
L4	"We'd ping the service from outside to check it's up."
L5	Designs it properly: "probers at multiple global vantage points, so we can tell a regional connectivity problem from a global outage."
Staff+	Bounds it honestly: "probing gives presence, not coverage — a few hundred vantage points against 100,000 autonomous systems, and the failures that matter are in networks we don't have a probe in. It's also synthetic, so it says nothing about a real user on a mobile network. Keep it, but it can't be the only mechanism."

Next: getting the signal from real users.