Home › Topics › “The Network Is Slow”: A Diagnostic Playbook › Chronic vs. Acute

Chronic vs. Acute Slowness: Two Very Different Investigations


A symptom is not a diagnosis. Medicine has known this for a long time. Chest pain that started five minutes ago and chest pain that has built for a decade point at different causes, get different tests, and end in different treatments. Network slowness works the same way. “The network is slow” describes two different diseases. The acute kind: things were fine yesterday and are bad today because something broke or changed. And the chronic kind: nothing broke at all, but something has been quietly growing until the network everyone remembers no longer exists.


Mixing them up is expensive in both directions, and I have paid both bills. Treat a chronic problem as acute and you hunt for a culprit that is not there. You reboot equipment, roll back innocent changes, and chase ghosts through logs while the real cause keeps growing. Treat an acute problem as chronic and you draft a budget request for new hardware to compensate for a failed cable or a bad configuration push. You buy your way around a defect that a twenty-minute fix would have cured.


This guide teaches you to tell the two apart quickly, mostly with one question and one graph. It then walks each investigation properly: the suspect lists, the tests, the fixes, and the different definitions of “done.” It closes with the mindset that stops chronic slowness from ever surprising you again: capacity thinking.


In this guide

  • One complaint, two diseases
  • Telling them apart: the timeline and the graph
  • The acute investigation: what changed?
  • The chronic investigation: what grew?
  • The trap cases
  • Capacity thinking
  • The two investigations, side by side
  • Where to go next

One complaint, two diseases


Acute slowness has a birthday. At some identifiable point — this morning, after the maintenance window, when the new firewall rule went in — performance stepped down. Before that moment, fine; after it, not. The cause is an event. It might be a failure: a link died and traffic squeezed onto a backup path, a port started taking errors, or a device is sick. Or it might be a change: config, firmware, a new application, or a new “temporary” box someone plugged in. Acute problems are found by asking what happened?


Chronic slowness has no birthday — only a trend. For example, there are ten more staff this year, files that doubled in size, and meetings that moved to video. Applications moved to the cloud and turned the internet link into the hallway everything walks through. Each week is imperceptibly worse than the last, until a complaint finally crystallizes. Nothing is broken; every device would pass inspection. The system has simply outgrown its capacity, or aged beneath its load. Chronic problems are found by asking what grew?


The reason the distinction matters so much is that each investigation's core move is wrong for the other disease. The acute move is to find and undo an event; there is no event behind chronic slowness to undo. The chronic move is to measure trends and add capacity or shed load. Adding capacity to route around a broken thing leaves the broken thing broken — and usually does not even help, since a defect is not a shortage.


A network can suffer both at once, and chronic strain makes acute problems louder. A link running near its ceiling has no headroom to absorb a misbehaving device that would pass unnoticed on an idle network. A switch at the edge of its abilities falls over from a burst it once would have shrugged off. When a “small” event produces outsized pain, that is often the chronic disease amplifying the acute one. That means you have two things to fix, not one. (The meeting about which to fix first is a third.)


Telling them apart: the timeline and the graph


Start with the question that costs nothing: “when did this start, and how suddenly?” Sharp answers (“Tuesday at ten”) lean acute. Foggy answers (“it's been bad for a while… worse lately”) lean chronic. But treat the answer as a lead, not a verdict, because memory is honest about the feeling and vague about the date. People adapt to slow decay without noticing — then one busy Thursday crosses their annoyance threshold and the complaint arrives sounding brand new. “It started Thursday” may honestly mean “Thursday is when I stopped tolerating it.”


The instrument that does not misremember is your monitoring history. This is the job the baseline exists for — the recorded picture of what normal has looked like over time. If the concept is new, what is a baseline is the place to start. Pull up months of history for the suspect resources — link utilization, latency probes, device CPU — and read the shape of the curve:


  • The step: flat, then an abrupt jump to a new level that persists. A before and an after. Acute — and the timestamp of the step is your best clue, because whatever happened, happened then.
  • The ramp: a curve drifting upward across weeks or months, no single inflection point. Chronic — growth in slow motion, and you can often extrapolate roughly when it will hurt in earnest.
  • The daily sawtooth: fine at night, painful at peak hours, every day. Chronic in nature — a resource that no longer fits the busiest hour — even though each afternoon's complaint feels like a fresh event.
  • The spiky mess: normal punctuated by irregular bursts. Neither disease exactly — an intermittent fault or a recurring heavy job. Timestamp the spikes and hunt what coincides with them.

The ticket queue itself is a second instrument worth reading. Acute problems produce a burst: several complaints inside one day, often naming the same thing, because many people crossed from fine to frustrated together. Chronic problems produce a trickle: one vague complaint this week, another next week, different people, different wording, each easy to dismiss individually. Glance back through recent tickets. You may find months of scattered “slow again today” grumbles you had been closing one by one. In that case, you are probably standing in a chronic investigation that has been open, unofficially, for a quarter.


No history to consult? Then today's job is twofold. Work the problem with the live-evidence techniques in the slowness evidence kit. Start collecting the history that will make the next investigation trivial. Every admin gets exactly one excuse on this; the second mystery slowdown with no graphs to consult is a choice.


The acute investigation: what changed?


Acute slowness is a detective story with a timestamp, and the suspect list is mercifully short. In rough order of likelihood, first is a change somebody made. That could be config, firmware, a patch, a new rule, or new equipment — by IT, a vendor, or a helpful colleague. Next is a partial failure: a link in a bundle down, a failed device quietly failed-over, or a port taking errors. This is the kind of degradation that slows things rather than stopping them. Next is a new load that arrived abruptly (a misbehaving machine, a runaway job, a new app rollout). Or it could be an upstream event at your ISP or a cloud provider. The sequence:


  • 1. Fix the moment of onset. From graphs if you have them (the step), from users if you don't. Everything else keys off this.
  • 2. Interrogate the change record. What did anyone touch in the day or two before onset? Check the change log, ticket queue, and chat — and ask out loud, blamelessly: confessions are diagnostic shortcuts. And a culture where “it might have been my change” is safe to say solves acute problems days faster.
  • 3. Sweep for silent failures. Device and interface status, error counters, redundancy state. A network limping on its backup path tells nobody unless monitoring is watching — loss and errors specifically are chased in the packet loss guides.
  • 4. Correlate the timestamp against everything: log entries, update schedules, the moment the new copier arrived. Coincidence in time is not proof, but it tells you where to look first.
  • 5. Undo or repair, one move at a time. Prime suspect identified? Revert the change or fix the failure — singly, so the recovery itself tells you which move cured it.
  • 6. Verify against the before. Done means metrics returned to their pre-onset baseline — not “seems better,” the same graph back at its old level.

A worked miniature: file access at a branch turns sluggish overnight. The latency graph to the branch shows a clean step at 2 AM. The change log shows nothing — but the firewall's logs note a reboot at 2 AM. And the firewall's interface now shows a fraction of its usual speed: after a power blip, one link renegotiated to a lower rate. Bounce the port, speed restores, graph steps back down. Twenty minutes — because the step's timestamp did the aiming.


One suspect deserves a special note: the upstream event. When the step in your graphs coincides with nothing you or your colleagues touched, the change may have happened outside your walls. Perhaps the ISP rerouted traffic, a provider had a rough morning, or a peering path shifted. You cannot see into their network, but you can bracket it. For example, your monitoring shows latency to the provider's edge stepping up while everything inside the building stayed flat. A quick test from a different network (a phone on cellular) shows the same destination behaving differently. That bundle — timestamped step, clean internals, corroborating outside test — is precisely what turns a support call from a plea into a report.


The chronic investigation: what grew?


Chronic slowness is not a whodunit; it is an audit. Nothing to revert, nobody to catch — instead you are looking for the resource that quietly ran out. The usual grown things include traffic (more people, richer applications, video everywhere, cloud services turning the internet link into the busiest cable you own). They include data (backups that no longer fit their window and bleed into the workday, files and databases that dwarf their old selves). They include density (a wireless design built for one laptop per person, now serving three devices each). And they include age (equipment from a smaller era, healthy but undersized for the load it now carries).


The method is comparison over time, not inspection in the moment. Take the busiest-hour view of each shared resource — internet and WAN links, uplinks, server resources, access points — and compare this quarter against earlier ones. You are hunting the resource whose peak periods now press against its ceiling, because congestion is a rush-hour phenomenon. A link can average half-empty and still ruin every afternoon. The arithmetic of that trap, and where the practical ceilings sit, is exactly the subject of how full is too full. Then identify what fills the resource — which conversations, which applications, which hours — before deciding the cure. Chronic fixes come in exactly three families. You can add capacity (upgrade the link, the uplink, the AP layout). You can shed or shift load (move backups off-hours, throttle the greedy, split traffic). Or you can accept and manage (prioritize what matters on the constrained resource). Which one is right is a judgment call about money, time, and how much the pain actually costs. That is why chronic investigations end in a proposal, not a fix ticket.


A worked miniature for this side too: a branch office reports the network is “worse every month,” with no event anyone can name. The branch link's history shows the ramp — busy-hour utilization drifting steadily upward across two quarters, no step anywhere. What grew? The branch gained staff and moved its phone system onto the network. Its nightly cloud backup quietly swelled until it now runs deep into the morning, colliding with the workday's video calls. Nothing to revert; three growths stacked on one link. The outcome is a package. It includes moving the backup window (shed load, free, today). It also includes a one-page proposal for the link upgrade the trend says is due anyway (add capacity, funded, next quarter). Both halves came from graphs, not guesswork, and the proposal was short because the ramp was attached.


Tip: in a chronic investigation, resist the word “fault.” Announcing that the aging core switch is “failing” when it is merely outgrown sends the conversation toward warranty claims and replacement panic instead of the real topic: capacity. Precise language — “healthy but undersized for what we now do” — gets better decisions and, oddly, faster budget approval. Decision-makers trust a measured trend more than an alarm.


The trap cases


Two disguises account for most misdiagnoses, and both are worth recognizing on sight. Trap one: chronic wearing acute's clothes. Utilization creeps up for a year, and nothing hurts — until it hits the zone where queues form, and then performance falls off a cliff in a single week. Users experience a sudden event (“it broke last Tuesday!”); the long graph shows the truth: a year-long ramp reaching a threshold. Queueing behaves nonlinearly — delay grows gently with load until, near saturation, it explodes. So the feeling of sudden onset is real even though no event occurred. The tell: the “step” in the pain has no corresponding step in the load; the load graph is a smooth ramp. Fix: capacity, not forensics.


At Ashgrove Housing the ticket said the network had broken on Tuesday, and for a week it was worked as an acute problem. A switch was rebooted, a firewall change was rolled back, and a vendor was called. Nothing had happened on Tuesday. The twelve-month graph, which nobody had opened, showed the internet link's busy-hour peak climbing steadily from half full to full. The queueing delay had arrived when that peak reached the top. Nobody had reported the year, because each week had been only slightly worse than the last. The housing officers had adjusted by starting large uploads before going for coffee. Tuesday was the day the adjustment stopped working.


Trap two: acute wearing chronic's clothes. Slowness seems to build gradually and diffusely — but the cause is a discrete defect. That could be a failing transceiver whose error rate climbs with temperature, or a duplex mismatch that punishes only heavy transfers. It could be redundancy that half-failed months ago so that everything now crowds one path. It feels like the network “wearing out,” and the reflex is to budget for upgrades. The tell here is error counters and asymmetry: growth does not produce CRC errors, retransmissions, or one path mysteriously idle while its twin runs hot. Healthy growth is clean; defects leave debris. Sweep for the debris before you draft the purchase request — the isolation tests in the four-zone method apply unchanged.


Capacity thinking


Acute problems will always ambush you; that is their nature. Chronic problems never should, because the evidence accumulates for months in plain sight — if anyone looks. Capacity thinking is the habit of looking: a short, recurring appointment with your trend graphs, asking three questions of every shared resource. Where is the peak now relative to the ceiling? Which way is it moving, and how fast? When, roughly, do the lines cross — and is that date closer than the lead time to fund and install more capacity? The moment the crossing date slides inside your procurement lead time, the upgrade conversation starts — calmly, with a graph, long before users feel anything.


Keep the output of that ritual as one small, living note — a capacity ledger anyone on the team can read at a glance:


Resource          Busy-hour peak   Direction     Status
Internet (down)   62% of ceiling   creeping up   fine; re-check monthly
Internet (up)     88% of ceiling   climbing      PROPOSAL DRAFTED
Core uplink       41% of ceiling   flat          fine
Branch link       74% of ceiling   climbing      watch; likely next
File server disk  busy at peaks    worsening     move backup window first

Five lines, five minutes a month, and the question “is this complaint chronic?” usually answers itself before the ticket is even filed. That is because you already knew which resource was next in line. The ledger is the least glamorous document you will maintain and the one that cancels the most meetings.


This is a modest ritual — monthly is plenty for most small networks, quarterly at minimum — and it converts chronic slowness from an emergency into a line item. It also quietly builds the historical record that makes the acute-versus-chronic question trivial next time: one glance, step or ramp, case sorted. If you are still choosing which handful of numbers deserve that recurring glance, the core network metrics guides lay out the short list.


The two investigations, side by side


AcuteChronic
Core questionWhat happened?What grew?
OnsetIdentifiable moment; a before and afterNo moment; a drift with a tolerance threshold
Graph signatureStep changeRamp, or worsening daily peaks
Prime suspectsChanges, partial failures, sudden new loadTraffic, data, device count, aging capacity
Key evidenceChange log, error counters, onset timestampLong-range trends, busiest-hour comparisons
Typical fixRevert, repair, removeAdd capacity, shift load, or manage it
TimescaleHours to daysWeeks to budget cycles
Done meansMetrics return to the old baselineHeadroom restored; trend re-checked on a schedule

Notice the last row especially: the two investigations do not even share a definition of success. Acute success is restoration — the old normal, back. Chronic success is a new normal with room to breathe, plus a habit that watches the next ceiling approach. Declaring victory the wrong way is its own trap. An acute fix verified against nothing tends to come back, and a capacity upgrade nobody re-measures gets eaten by the same growth that forced it.


Where to go next


One question, one graph: how suddenly did it start, and does the history show a step or a ramp? Get that right and everything downstream — suspects, tests, fixes, even the tone of the conversation — falls into place. Get it wrong and you will be rebooting a switch to cure a year, or budgeting for a cable. The acute path leans on the intake and isolation guides earlier in this series, beginning with your first ten minutes. The chronic path leans on baselines and trend review. Both live with the rest of the playbook at “The Network Is Slow”: A Diagnostic Playbook.


Put this into practice with SysPeep Monitor

The step-or-ramp question has a five-second answer when the history already exists. SysPeep Monitor collects utilization, latency, errors, and device health around the clock over SNMP and WMI. It keeps the long-range graphs, and lets you set thresholds that flag a creeping trend before it becomes a Thursday emergency. The free trial installs on any Windows machine.

Download Free TrialLearn About SysPeep