Home › Topics › Path & Hop-by-Hop Analysis › Building Path Baselines

Know Your Normal: Building a Path Baseline


Every troubleshooting guide eventually tells you to run a traceroute. This is the command-line tool that lists the routers, or hops, your packets cross on the way to a destination. Very few guides mention the awkward part. A traceroute captured during a problem is nearly worthless on its own, because you have nothing to compare it with. Is hop four always that slow? Did the path always run through that provider? Was it always eleven hops long? Without a record of normal, you end up squinting at a wall of addresses and milliseconds, seeing whatever your current suspicion wants you to see.


A path baseline fixes that. It is a small, honestly rather boring collection of records. It records the routes your traffic takes to the destinations that matter, captured on good days and saved somewhere you can find them on bad ones. Once it exists, you stop guessing and start comparing. This guide walks you through building one. It covers which destinations to cover, what to write down, and how to automate the capture with scheduled traceroutes. It also covers how to keep the whole thing truthful as your network changes underneath it.


In this guide

  • A baseline is evidence, not a diagram
  • What "normal" actually includes
  • Choosing destinations worth baselining
  • Recording internal paths
  • Recording external paths
  • The path register: one page per destination
  • Automating the capture
  • Keeping the baseline honest
  • Where to go next

A baseline is evidence, not a diagram


You probably already have a network diagram, and if you do not, you should. But a diagram and a baseline answer different questions. The diagram documents intent: traffic from the office to the branch is supposed to ride the private WAN circuit. The baseline documents observed reality. For example, on a healthy Tuesday afternoon, a trace from a sales PC to the branch file server took six hops. It entered the provider network at a specific address, and landed in about twenty milliseconds. Intent and reality drift apart more often than anyone likes to admit, and the gap between them is frequently where your problem lives.


Think about your commute. You notice a detour within seconds, because you drive the same road every day and your brain holds an effortless baseline of it. You have no such instinct for network paths you only look at during a crisis. The first time many administrators ever trace the route to their most important application is the day it breaks. That means they are learning what the road looks like and judging whether it is broken at the same time, under pressure, with a manager hovering.


A recorded baseline earns its keep in three situations. During an incident, it turns "this trace looks bad, I think?" into "hops one through four match the baseline, the latency jump at hop five is new". That is the starting point of the localization playbook. Between incidents, it is the reference that automated path change detection compares against. And when you finally open a ticket with your provider, a before-and-after pair of traces makes the difference. It separates being taken seriously from being asked to reboot your modem.


What "normal" actually includes


A baseline is more than a stack of raw trace output. "Normal" has several dimensions, and each one catches a different kind of change later:


ElementWhat to recordWhy it matters
Hop countA range across several runs, not one numberA detour usually changes the length of the road
The hop sequenceThe address of each router, plus its DNS name when one existsThis is the road itself — the thing you will diff
Per-hop latencyA rough range for each hop, sampled at different times of dayRanges beat single numbers; busy-hour and quiet-hour differ
End-to-end latencyThe range to the final destinationThe only number users actually feel
Handoff pointsWhere your equipment ends and each provider beginsAssigns ownership instantly when something breaks
Provider namesThe domains appearing in the routers' DNS namesA new provider in the path means your traffic was rerouted
Known variationHops that legitimately alternate between runsSo routine variation does not trigger panic later

Two of these deserve a closer look. First, latency: every figure a traceroute prints is a round-trip time, the full out-and-back journey. That is why it varies from run to run and why you record ranges instead of single readings. The guide to the anatomy of a round trip explains what hides inside that number. Second, the DNS names: routers often carry descriptive reverse DNS names — the name you get when you look up an address. Those names usually contain the provider's domain. They are how you will later notice that your packets quietly switched carriers.


One caution before you fill in the per-hop columns: individual hops have measurement quirks. Routers deprioritize the replies traceroute depends on, so a single hop can look slow or lossy while forwarding your actual traffic perfectly. Record what you see, but read it with the interpretation habits from reading traceroute output before treating any one hop's number as gospel.


Choosing destinations worth baselining


You cannot baseline the internet, and you should not try. The goal is a short list — five to ten destinations — chosen so that when one of them misbehaves, the phone rings. Good candidates share a few traits. The business depends on them, or they are sensitive to latency or loss. The path to them may cross a boundary you do not control. Or they simply have a history of generating tickets.


A sensible starter list looks like this:


  • User segment to critical servers. Start from a machine in a typical user VLAN — a virtual LAN, one of the logical segments your switches divide the network into. Trace to the file server, the directory server that handles logins, and the main line-of-business application.
  • Server segment to its dependencies. From the server network to the backup target, the replication partner, or the database the application leans on.
  • Office to each key SaaS. The hosted CRM, the mail service, the accounting platform — whatever the company would riot without.
  • Office to each branch or VPN peer. Site-to-site paths break in their own special ways, and each direction deserves its own entry.
  • Office to your DNS resolvers. When the resolver path degrades, everything looks broken at once.

Tip: pick one stable, representative source machine per segment and always test from it. A trace is only meaningful together with its starting point. A baseline collected from the server room tells you nothing about the road a sales PC travels.


Recording internal paths


Internal paths are short. Between a user PC and a server there are usually only two to four routed hops. That is because most of the journey happens inside switches that never appear in a trace. Here is a typical internal capture, taken from a PC in the user VLAN toward a file server, with annotations added:


$ traceroute -n 10.40.8.25
 1  10.20.0.1   0.4 ms   <- user VLAN gateway (core switch)
 2  10.0.1.2    0.8 ms   <- datacenter firewall
 3  10.40.8.25  1.0 ms   <- the file server itself

Normal, for a wired internal path, is strict: latency from a fraction of a millisecond to a few milliseconds, and no loss at all. There is no distant provider to absorb the blame inside your own walls. So any sustained internal loss in a baseline capture is not a quirk to note down. It is a fault to fix before you declare this "normal." Baselining has a pleasant side effect of surfacing these embarrassments early, while nobody is watching.


Notice what the trace does not show. That single hop between the PC and its gateway may physically cross an access switch, an uplink, and a distribution switch. All are invisible here because they forward traffic without ever decrementing the counters traceroute relies on. For the tools that see into that switched layer, read the companion guide on internal versus external path analysis.


Capture the internal set from each user segment that matters — the office VLAN, the warehouse VLAN, the wireless network. Their gateways and firewall paths can differ, and so can their problems.


Recording external paths


External paths are where baselines really pay off, because most of the road belongs to someone else and changes without notice. Here is an annotated capture from the office toward a hosted CRM:


$ traceroute crm.example.net
 1  fw1.corp.example (10.0.1.2)                 0.6 ms
 2  198.51.100.1                                2.1 ms   <- ISP handoff: your circuit
 3  ae3.cr1.example-isp.net (198.51.100.77)     3.4 ms
 4  ae9.br2.example-isp.net (198.51.100.130)    9.8 ms
 5  xe0.peer.transit.example (203.0.113.5)     10.2 ms   <- handoff to transit carrier
 6  203.0.113.61                               82.0 ms   <- long-haul jump: normal here
 7  edge2.hosting.example (203.0.113.9)        84.1 ms
 8  crm.example.net (203.0.113.30)             85.0 ms

Read it the way a future incident will need it read. Hop two is where your network ends — the first address that belongs to your internet provider. Hops three and four carry the provider's domain in their names. Hop five hands off to a transit carrier, a wholesale network that hauls traffic between providers. The eighty-millisecond jump at hop six is geography — a long-haul link — and it is normal for this destination. Write that down explicitly. Six months from now, someone troubleshooting under pressure will discover that jump and gasp. They will chase it for an afternoon unless the register says it has always been there.


Run the capture several times before you call it a baseline. Providers commonly spread traffic across parallel links using ECMP (equal-cost multipath). So consecutive runs may alternate between sibling routers in the middle of the path. That alternation is part of normal — record the alternates as known variation instead of letting them surprise you later. A long-running MTR session repeats the path test continuously and summarizes per-hop statistics. It is the comfortable way to see this variation in one screen. See MTR path testing for the tool itself.


While you are capturing, spend two extra minutes on dimensions a traceroute alone does not cover. Loss is the big one: fire a few dozen pings at the destination and write down the result. That is because "zero loss end to end on a healthy day" is exactly the sentence you will want in the register when a bad day shows three percent. The address family is the other. If a destination resolves to both an IPv4 and an IPv6 address, you do not have one path to it. In that case, you have two, and the operating system quietly picks one per connection. Trace each explicitly (most tools accept -4 and -6 flags) and give each its own baseline entry. That way, the day one of them degrades you are not comparing a sick path against a road you never mapped.


One practical habit: capture with name resolution turned off for speed and consistency. But keep an occasional resolved run too, because the DNS names are where the provider identities live. You want both forms in the archive.


The path register: one page per destination


Raw trace files are your evidence; the path register is the readable summary a stressed human turns to first. It is one short entry per critical destination. A wiki page, a text file in a repository — anything the whole team can reach when things are on fire will do. An entry looks like this:


FieldExample entry
Destinationcrm.example.net — the hosted CRM
Why it mattersSales floor stops working without it
Test sourcemon01, server VLAN (plus spot checks from a sales PC)
Expected egressFiber circuit via fw1; ISP handoff at 198.51.100.1
Expected providersexample-isp.net, then transit.example
Hop countEight to nine
End-to-end latencyRoughly 80 to 95 ms
Known variationHops three and four alternate between cr1 and cr2
Last verifiedDate and initials of whoever last re-ran the capture

Every field earns its place, but the last one is load-bearing. An undated baseline slowly turns into a lie: the network changes, but the register does not. And the next person to trust it gets sent confidently in the wrong direction. Date the entry, and re-date it every time someone verifies it.


Resist the urge to make the register fancy. Plain text survives outages, prints well, and diffs cleanly in version control. The register you can read while the wiki server is unreachable is worth two that you cannot.


Automating the capture


Nobody remembers to run traceroutes on a schedule, so do not rely on remembering. A tiny script plus the system scheduler builds your archive while you sleep. On a Linux box or a monitoring probe, something like this run hourly from cron does the job:


# capture-paths.sh - run hourly from cron
STAMP=$(date +%Y%m%d-%H%M)
DIR=/var/lib/pathbase
for HOST in fs01.corp.example crm.example.net 198.51.100.25; do
  traceroute -n -w 2 -q 2 "$HOST" > "$DIR/$HOST-$STAMP.txt" 2>&1
done

On Windows, the built-in tracert plus Task Scheduler gets you the same archive:


# capture-paths.ps1 - run hourly from Task Scheduler
$stamp = Get-Date -Format yyyyMMdd-HHmm
foreach ($h in "fs01.corp.example","crm.example.net") {
  tracert -d $h | Out-File "C:\PathBase\$h-$stamp.txt"
}

A few practical notes. Hourly is plenty for baseline-building; these are text files, so a deep archive costs effectively nothing — keep months of history. Name files by destination and timestamp so you can grab "the same path, last Tuesday, same hour" in seconds. Run the job from the machines your register names as test sources, because a capture from a different vantage point is a different baseline. And keep the flags gentle — one or two probes per hop with a short timeout — since this runs forever against production paths.


Scheduled traceroutes record the road; they are a thin sample of the ride. Two probes an hour says very little about what latency did between captures. So pair the archive with a continuous latency grapher. This is a tool that pings each destination many times a minute and draws the loss and the spread over weeks. SmokePing-style latency graphing fills exactly that gap. Its long-term charts are the honest source for the latency ranges in your register. Read the range off the graph instead of guessing it from a handful of trace lines.


Notice what the script deliberately does not do: compare, judge, or alert. That restraint is temporary. Once a few weeks of captures exist, the natural next step is to diff each new capture against the baseline automatically and get told when the road changes. That is exactly the subject of detecting path changes.


Keeping the baseline honest


Baselines rot. Circuits get replaced, firewalls get swapped, providers reroute, and a register nobody maintains quietly becomes fiction. The fix is to bolt baseline review onto the moments when the network actually changes. Review after any change that could touch a path. Such changes include a new circuit, a firewall replacement, a routing adjustment, a VPN redesign, or a provider migration. Walk this list:


  • 1. Re-run the captures for every register entry the change could plausibly affect, right after the change window.
  • 2. Compare against the register. Hop count, providers, latency ranges — not just whether the destination still answers.
  • 3. If the difference was intended, update the entry, note the date, and move on with a clear conscience.
  • 4. If the difference was not intended, investigate before closing the change. An unexplained new path minutes after maintenance is the cheapest early warning you will ever get.
  • 5. Glance at performance, not just topology. The path can look identical while its latency quietly doubled.

Between changes, put a recurring reminder on the calendar to re-verify the register even though "nothing changed." Every few months is a reasonable rhythm. Providers re-engineer their networks on their own schedule. Your traffic can be taking a new road today without a single ticket in your change system mentioning it.


Tip: make the review a named step in your change checklist — "re-verify affected path register entries" — rather than a good intention. Checklists survive busy weeks; intentions do not.


Where to go next


You now have the whole discipline in hand. You have a short list of destinations that matter and scheduled captures quietly filling an archive. You have a plain-language register summarizing normal for each path, and a review habit that keeps it all truthful. None of it is glamorous, and all of it compounds. Every future path problem starts from "compare with the baseline" instead of from zero.


From here, one natural continuation is automating the comparison in detecting path changes before users notice. Another is using your baseline under fire in finding the slow hop. Then there is learning why the return road may not match your outbound map in asymmetric routing. For the broader craft of judging what latency numbers should look like in the first place, the network latency guides are the companion shelf. The rest of the Path & Hop-by-Hop Analysis pillar builds directly on the normal you just recorded.


Put this into practice with SysPeep Monitor

A path register tells you what the road normally looks like; monitoring tells you the moment the ride changes. SysPeep Monitor watches latency, packet loss, and interface health across your devices around the clock. It has thresholds and alert rules tuned to stay quiet until something real happens. Pair it with your new baseline and the free trial installs on a Windows machine in minutes.

Download Free TrialLearn About SysPeep