Home › Topics › Path & Hop-by-Hop Analysis › Internal vs. External Paths

Internal vs. External Path Analysis: Two Different Games


Run a traceroute to your file server and you get three tidy hops. Run one to a cloud service and you get fourteen messy ones. Same command, same screen — but you have just switched between two fundamentally different games. Playing one with the other's rules is how path analysis goes wrong. Inside your walls, the trace shows almost nothing (most of the journey happens in devices it cannot see). But you own every inch and can fix anything. Outside, the trace shows almost everything, and you can fix approximately none of it.


This guide teaches both games properly. For the internal game, it explains why the routed path is only the skeleton and how to trace the switched path that traceroute is blind to. It covers the interface counters that act as your per-hop health data. For the external game, it explains how to interpret a road you will never control and what actually belongs to you out there. It shows how to escalate to a provider with evidence that gets acted on instead of politely ignored. And because every real incident starts with "which game am I even playing?", there is a decision guide per zone.


In this guide

  • Same command, two different games
  • The scoreboard: internal versus external at a glance
  • The internal game: three hops and a big blind spot
  • Tracing the switched path traceroute cannot see
  • Reading the switched path's health
  • The external game: reading a road you cannot fix
  • Escalating to your provider with evidence
  • Picking the right game per zone
  • Where to go next

Same command, two different games


The dividing line is ownership, and it sits at your internet handoff — the point where your equipment ends and the provider's begins. Everything on your side is the internal game: few routed hops, total access, strict standards, and the luxury of walking over and touching the misbehaving box. Everything past the handoff is the external game: many hops, zero access, elastic standards that depend on geography, and influence limited to what you can prove.


The reason this matters is that each game punishes the other game's habits. Bring external patience to an internal problem and you will shrug at loss you should be hunting down. "paths are just like that sometimes" is never true inside your own walls. Bring internal entitlement to an external problem and you will burn hours trying to "fix" a distant router that is not yours, not broken, and not listening. Know which side of the handoff the problem lives on. That is exactly what the localization playbook determines. Then play that side's game.


The scoreboard: internal versus external at a glance


DimensionInternal gameExternal game
Routed hops in a traceA handful — often two to fourMany — often eight to twenty
Who owns the gearYou, end to endYour provider, then strangers
What the trace hidesEvery switch — most of the actual devicesLittle — but middle hops answer lazily
Healthy latencySub-millisecond to a few millisecondsWhatever geography dictates; judged against baseline
Sustained lossZero, full stopZero end to end; single-hop blips are often artifacts
Core toolsInterface counters, MAC tables, neighbor discoveryMulti-vantage traces, continuous stats, the path register
Your powerFind it and fix itProve it and escalate it
Typical timescaleSame dayA ticket and some patience

Print that table into your head and half the judgment calls in path work make themselves. The rest of the guide is the two columns in detail.


The internal game: three hops and a big blind spot


Why does a trace across your whole building show only three hops? Because traceroute counts routers — devices that move packets between different networks and decrement the packet's hop counter as they do. Your network has only a few of those: a core switch doing the routing between VLANs, maybe a firewall between zones. The rest of the journey rides through switches operating at layer two — the Ethernet layer. These forward frames within a network segment by MAC address. They never touch the hop counter, and are therefore completely invisible to traceroute.


Concretely: the "one hop" between a sales PC and its gateway might physically be PC → wall jack → access switch in the closet → uplink → distribution switch → core. Five cables and three switches, any of which can be the problem, all compressed into a single innocent-looking line of trace output. That compression is the internal game's defining feature: the routed path is a skeleton. The switched path is the flesh; and your per-hop tools only X-ray the skeleton.


So the internal game is played in two moves. Move one: use the routed trace to pick the segment. If PC-to-gateway is where the latency or loss appears, the problem lives somewhere in that switched stretch. Move two: reconstruct the invisible stretch with the switch's own bookkeeping, then check the health of every port along it. The next two sections are those moves in detail.


One internal wild card before the moves: wireless. An access point is just another invisible layer-two device on the switched stretch. But the air between it and the laptop is the single segment inside your walls where the strict zero-loss standard does not apply. Radio interference, distance, and crowding all legitimately cost packets. So when the complaining machine is wireless, split the question immediately: repeat the test from a wired machine on the same VLAN. Wired clean and wireless ugly means the air, the AP, or the client's radio — a different hunt entirely. Both ugly means the switched path behind the AP, and the moves below apply unchanged.


Tracing the switched path traceroute cannot see


Switches may be invisible to traceroute, but they are not secretive. Two kinds of bookkeeping reveal the path. Neighbor discovery protocols (most managed switches speak one) let each switch announce itself to its direct neighbors. Reading the neighbor table tells you which switch connects to which, drawing the map one link at a time. And the MAC address table records, for every MAC address the switch has seen, which port leads toward it. That table is your hop-by-hop tool for the switched world: ask each switch "which way to this machine?" and follow the pointing fingers.


The commands vary by vendor, but the shape is universal. Say the slow machine is 10.20.8.31 and you are hunting the physical path from the core to its desk:


# 1. On the gateway: translate the IP address to a MAC address
core1# show ip arp 10.20.8.31
  10.20.8.31    001a.2b3c.4d5e    Vlan20

# 2. Ask the core: which port leads toward that MAC?
core1# show mac address-table address 001a.2b3c.4d5e
  20   001a.2b3c.4d5e   DYNAMIC   Gi1/0/48     <- an uplink, to closet2

# 3. Follow the uplink and ask the next switch
closet2# show mac address-table address 001a.2b3c.4d5e
  20   001a.2b3c.4d5e   DYNAMIC   Gi0/17       <- an access port: the desk

Each answer points one link farther along the road. The MAC lives out port 48, which your documentation (or the neighbor table) says is the uplink to the closet switch. The closet switch says port 17, which is a wall jack. The chain you just walked — core Gi1/0/48 → closet2 → closet2 Gi0/17 → desk — is the switched path, reconstructed in three commands. Traceroute compressed it to one hop; the MAC tables gave every device and port on it a name.


If a lookup comes back empty, do not panic — these tables are caches, not ledgers. An ARP entry or a MAC entry that has not been used for a while quietly expires. A machine that has been idle can vanish from every table between you and it. The fix is charmingly circular. Ping the machine once, which forces fresh traffic through the whole stretch, repopulating each table along the way. Then run the lookups again while the trail is warm. And if the entry flaps — the same MAC appearing on different ports between readings — stop and pay attention. That is either a machine genuinely moving or a switching loop wearing a disguise.


Tip: this walk is fast when interface descriptions are filled in ("uplink to closet2", "jack 2-14") and miserable when they are blank. Every minute spent labeling ports and keeping the wiring map current is repaid the first time you chase a MAC at speed with a manager watching.


Reading the switched path's health


Now that the invisible stretch has names, you can ask each named port how it feels. Interface counters — the per-port tallies every managed switch keeps — are the internal game's equivalent of per-hop statistics. They include errors, discards, utilization, and collisions, one set per port, updated live. Walk the ports along your reconstructed path and the guilty link usually confesses in numbers:


Symptom on the segmentClassic culpritThe counter that tattles
One machine slow, transfers stall and recoverDamaged cable, patch lead, or jackInput errors / CRC errors climbing on its access port
Sluggish and weird under any real loadDuplex mismatch — the two ends disagree about full- versus half-duplexLate collisions on one side, errors on the other
Whole closet slow at busy hours, fine at lunchSaturated uplinkOutput drops on the uplink port, utilization near line rate
A VLAN periodically freezes solidA loop or broadcast stormBroadcast counters exploding, switch CPU pinned

Two habits make counters trustworthy. First, read them twice. A big error number proves only history. A number that grows between two readings minutes apart proves an active problem. Second, remember the strict standard of the internal column. On a healthy wired LAN, error counters sit still and loss is zero. There is no "eh, some errors are normal" inside your own walls. A counter that climbs has a physical cause with a findable location. And the fix is usually a cable, a transceiver, a port, or a configuration line, applied the same afternoon. This is also precisely the data a monitoring system collects from every port around the clock. That turns the whole hunt into "look at which graph went red". But the manual walk above is what makes those graphs mean something to you.


The external game: reading a road you cannot fix


Past the handoff, the toolset inverts. Traceroute now shows you plenty — a dozen or more routers with names and numbers. Your job shifts from fixing to interpreting, because none of them will ever accept your login. Three reading disciplines keep the interpretation honest. Middle hops that look lossy or slow while later hops run clean are usually routers deprioritizing measurement replies, not real trouble. This is the classic trap unpacked in reading traceroute output. Hops that alternate between runs are load-balanced parallel links, part of normal. And a big latency jump mid-path is often just geography — an ocean happening. That is why every external judgment leans on the recorded normal from your path baseline rather than on gut feeling. Add the structural caveat that the return half of every external road is invisible from your side. This is the subject of asymmetric routing. Then you have the full reading kit.


The far end of the road deserves its own sentence in the rulebook. The last hop or two belong to the destination — its host, its provider. Trouble may sit there, seen identically from every vantage point including ones outside your network. In that case, you have localized a problem you cannot even file a circuit ticket about. Your moves shrink to: check the service's status page, report what you measured, and manage expectations at home with the evidence in hand. Knowing that this outcome exists — and being able to prove it in twenty minutes instead of guessing all afternoon — is itself one of the external game's wins.


What do you actually own out there? Three things. The handoff itself: continuous tests to the first provider address measure your circuit in isolation. A long MTR session toward it is the cleanest circuit-health check you can run without a truck roll. Your egress equipment: the edge router or firewall whose counters and utilization graphs rule your own driveway in or out. This is always checked before blaming anyone, because a saturated uplink imitates every provider fault beautifully. And your evidence: baselines, captures from several vantage points, timestamps. In the external game, evidence is not documentation; evidence is the only lever you hold.


Escalating to your provider with evidence


Escalate when localization puts the trouble at or beyond the handoff, it persists, and it hurts real work. The difference between a ticket that gets escalated to engineers and one that dies in the first-line script is almost entirely in how you package what you know:


  • 1. Lead with the circuit ID and the handoff addresses from your register. It routes your ticket to the right team and signals you know your own edge.
  • 2. State the impact in numbers: loss percentage, latency now versus your baseline range, since when — with timestamps and your time zone.
  • 3. Attach before and after: a healthy-day capture and a current one, same source, same options. A diff is worth a thousand adjectives.
  • 4. Show more than one vantage point: "seen identically from two internal segments; a vantage outside your network reaches the same destination cleanly" is a sentence that ends debates.
  • 5. Use continuous statistics, not a single screenshot — a summarized run of a few hundred cycles carries actual weight.
  • 6. Say what you ruled out: edge utilization normal, interface counters clean, tested past your own equipment. You are answering their first three questions in advance.
  • 7. Ask for something specific: circuit tests, error readings on their side of the handoff, recent maintenance or rerouting affecting your path.

Then expect the game's rhythm. They will test from their side. Their vantage point sees their half of the road in ways yours cannot. A fair amount of real provider troubleshooting is the two of you triangulating across the fence. Stay factual, re-send measurements as they change, and keep your own captures running through the whole episode so recovery is as well-documented as failure.


And close the loop like an engineer. When the provider declares the issue resolved, verify against your baseline before you agree. Latency should be back inside the register's range, and loss back to zero. The path itself should be back to its golden shape or consciously re-blessed as a new normal. "The alarm cleared" and "the road is healthy again" are different claims, and you now have the instruments to tell them apart. A one-line note in the register — what broke, where, how long, ticket number — turns this incident into calibration for the next one.


Warning: never open the ticket with a screenshot of one traceroute showing one scary middle hop. First-line support sees a hundred of those a week, almost all artifacts. Being wrong about it once costs you credibility on every future call. Send the persistent, multi-vantage, baseline-compared story instead.


Picking the right game per zone


Real problems rarely announce their zone, so start every path investigation with the split:


  • Destination inside your network? Pure internal game: routed skeleton first, then the switched path and its counters.
  • Destination outside? Two games in series, divided at the handoff. Localize first. The playbook tells you which side of the line the trouble lives on. Then play only that side's game: counters and cables inward, evidence and escalation outward.
  • Site-to-site over a VPN? The sneaky hybrid. The tunnel behaves like an internal link and shows up as one hop. But that hop rides a full external road underneath, invisible the same way switches are invisible inside a routed hop. Test both layers. Test through the tunnel (the overlay) like an internal path. Test to the far site's public address (the underlay) like an external one. When the tunnel is "slow," the underlay test tells you whether the problem is the road or the tunneling itself.

The hybrid case generalizes into the guide's closing thought. At every scale, path analysis is the same two-step. Find the visible hop that contains the problem, then open it up with the tools of whatever layer hides inside. Traceroute compresses switches into routed hops and whole internet roads into tunnel hops; the skill is knowing which magnifying glass each kind of hop deserves.


Where to go next


You now hold both rulebooks: inside — skeleton trace, MAC-table walk, counters, strict standards, fix it today; outside — baseline-anchored interpretation, multi-vantage proof, disciplined escalation, patience. Between them sits the handoff, the most important address in your register.


The companion guides complete the kit. The guide to building path baselines records normal for both zones before you need it. The guide to detecting path changes watches the external roads that shift without asking. For the full shelf, head back to Path & Hop-by-Hop Analysis.


Put this into practice with SysPeep Monitor

The internal game runs on interface counters — and watching every port by hand does not scale past one closet. SysPeep Monitor auto-discovers your switches and interfaces. It collects errors, discards, and utilization over SNMP around the clock. It alerts you when a counter starts climbing. It has live maps that show exactly which link is hurting. The free trial installs on a Windows machine in minutes.

Download Free TrialLearn About SysPeep