Home › Topics › Top Talkers: Who's Eating the Bandwidth › You Found the Hog. Now What?

You Found the Hog. Now What?


The technical hunt ends with a name on your screen: this machine, this application, this many gigabytes. It feels like the finish line. It is not, because the machine has a person attached and the application has a purpose attached. What happens in the next hour decides how this incident ends. It could end with a better network or with a colleague who flinches every time IT walks past their desk.


This is the part of the job no certification covers, and it is where junior admins step on the most rakes. I stepped on both of the big ones in my first year. The first is marching over with an accusation and being wrong, publicly, about a backup client the user has never heard of. The second is silently blocking the traffic and walking away, then discovering next week that you broke the thing a whole department depended on. Both mistakes cost the same currency, trust, and trust is what makes the next incident easy or impossible.


This guide is the playbook for the human step. It covers how to verify before you approach anyone and how to open the conversation so it stays a conversation. It explains the four standard resolutions and when each fits, plus what to do when the hog is the business. It also shows how to write the outcome down so one incident buys you a permanent improvement.


In this guide

  • You found a device, not a culprit
  • Opening the conversation
  • The four resolutions: reschedule, throttle, relocate, block
  • When the hog is the business
  • Write it down
  • From one incident to better policy
  • The conversations that go wrong

You found a device, not a culprit


Repeat this until it is reflex: flow data identifies machines, not people. Flow data is the per-conversation traffic record the hunt was built on, and the most it can say is which address moved the bytes. The distance between “this workstation moved forty gigabytes” and “this person did something wrong” is wide, and most incidents live in that gap. The machine may be shared — a conference-room PC, a shop-floor kiosk, a hot desk. The traffic may be automatic: a sync client, a backup agent, or an update service doing its job with no human involvement whatsoever. That software may even have been installed by your own team. The user may be logged in but absent while software acts on their behalf. Or the machine may be compromised, in which case its owner is the first victim, not the offender.


So before approaching anyone, spend five minutes answering four questions. Skipping them is how the quiet person in accounting gets blamed for a conference-room PC that was streaming a webinar.


  • Is the traffic automatic or human-driven? The application you identified usually answers this — sync clients, backup agents, and updaters run themselves. If the answer is “automatic,” there may be no behavior to discuss at all, just a setting.
  • Who actually uses this machine? Check the inventory, the logged-in account, and whether it is a shared or personal device. Guessing wrong here is how the wrong person gets blamed for a shared machine's traffic.
  • Does the timing match a person at all? Traffic at hours when the office was empty points at scheduled software (or something worse), not the name on the asset tag.
  • Could this be compromise? Run the traffic through the signals in classifying bandwidth hogs first. If malice is genuinely on the table, this stops being a friendly chat and starts being your security process — and tipping off the user may be exactly wrong.

One more check before you stand up: glance at the graph again and see whether the traffic is still running. If the burst ended an hour ago, there is no emergency to interrupt anyone over. The conversation becomes a relaxed “for next time” chat, which is both easier to have and easier to hear. Matching your urgency to the graph's urgency is part of getting the tone right.


A wrongly accused colleague remembers it for years, and they retell it. One public misfire and people stop reporting problems to IT, start hiding workarounds, and your visibility into your own network gets worse. Five minutes of verification is the cheapest insurance you will ever buy.


At Ashgrove Housing a new admin walked up to a housing officer's desk with a printout and opened with “your machine is the problem.” The officer had spent the morning uploading inspection photos because the case system required them by noon. And she had done nothing wrong. She said very little that afternoon and rather more to her manager the next day. A week later the senior admin went back, sat down, and started again with “can you help me understand what this does for you?” The answer took four minutes and contained the whole fix. The photo app uploaded full-size originals, had a setting for smaller ones, and nobody had ever shown her where it was. She now messages IT the day before an inspection run. The printout was not mentioned again.


Opening the conversation


Assuming a human is worth talking to, the frame that works is curiosity, allied with them. You are not delivering a verdict; you are investigating together. The honest framing — and it genuinely is honest — is that you are trying to make their work and the network both succeed. “I'm trying to help this app do its thing without flattening everyone's video calls.” You come with evidence, questions, and options. You do not come with a printout, a folded-arms stance, and a policy citation.


The wording does a lot of the work:


Instead of…Try…
“Your machine is killing the network.”“I'm chasing a big traffic spike and the trail leads to this machine — can you help me figure out what it's doing?”
“Did you upload something huge this morning?”“Did anything new happen with files or apps this morning — big folder, new tool, anything syncing?”
“You can't use that app here.”“That app is moving more data than our link can absorb during work hours — let's find a way it can run without hurting everything else.”
“This violates policy.”(Save policy for the follow-up, in writing, if it is needed at all. Leading with it ends the information flow.)

Prefer voice or a desk visit over email for the first contact — tone survives speech far better than text. And a two-line email reading “we need to talk about your network usage” will ruin someone's morning in a way you did not intend. Ask, then listen: the user's answer usually contains the diagnosis. “Oh, the vendor gave me a drive of footage and I copied it into the shared folder.” Frequently they had no idea anything was wrong, because from their side, nothing was. The app accepted the files; the app said nothing about a shared pipe. Almost nobody hogs bandwidth on purpose. Treat them accordingly and they become your early-warning system next time. There is no better network sensor than a user who trusts you enough to say “hey, I'm about to move something huge — when's a good time?”


If it helps to have a script, the conversation has three beats. Show: one sentence of what you observed, in plain terms (“the office internet was maxed out this morning, and the traffic traces back to this machine”). Ask: what changed, what is new, what runs on this box — then actually listen, because this is where the diagnosis usually finishes itself. Propose: solve it together, out loud, with the options in the next section. Three beats, five minutes, no printouts slapped on desks.


The four resolutions: reschedule, throttle, relocate, block


Nearly every hog resolution is one of four moves, or a combination. Which one fits depends on the verdict from classification and on one question. Ask it out loud: does this traffic need to happen, and does it need to happen now?


Reschedule — right job, wrong hour

The workhorse. The traffic is legitimate but it is competing with humans for daytime bandwidth. And it does not actually care what time it runs — backups, replication, big sync jobs, bulk downloads rarely have opinions about o'clock. Move it to a window when the pipe is idle, and confirm the whole job fits the window. A backup that starts at midnight but finishes at ten in the morning has only moved the problem. Pick the window from evidence, not folklore. The link's overnight utilization graph shows you which hours are genuinely dead. It also shows whether every other rescheduled job has already been piled into the same popular midnight slot. Stagger them. Then check the graph the morning after the first run to confirm the job behaved. Cheapest fix, least friction, and the one to try first whenever the job is automatic.


Throttle — right job, wrong appetite

When the traffic must run during shared hours, cap its rate. There are two places to do it, in order of preference. First, in the application: most backup, sync, and transfer tools have a bandwidth-limit setting. An app-level limit is precise, visible to the app's owner, and impossible for the network to misapply. Second, on the network path: use your router or firewall's rate-limiting or shaping features against that machine or service. These settings cap how fast a given machine or service may send. Network-side throttling is the fallback for apps with no limiter and for hosts you do not control. Size the cap so the link stays comfortable even when the job runs at full cap during the busiest hour. A minority share of the pipe is the right instinct. If the tool supports schedules, let it run capped by day and unthrottled overnight: the best of both resolutions. A cap set this way turns a monopolist into a polite tenant: the job still finishes, just not at everyone else's expense.


Relocate — right job, wrong geography

Some hogs exist because data is taking an absurd route. Thirty machines each downloading the same update across the internet link is a relocation problem. Stand up a local update server or enable peer caching, and the bytes cross the pipe once. Peer caching means machines fetch the update from each other instead of from the internet. A branch office hammering the site-to-site link for files everyone needs daily is a relocation problem: replicate the data to the branch overnight. A team round-tripping media through a cloud service to move it between two desks in the same building is a relocation problem with a punchline. The question that finds these: does this data have to travel this path at all?


Block — rare, and never a solo decision

Blocking — denying the traffic outright at the firewall or on the machine — is the right answer far less often than it is the tempting one. It fits three cases. The first is malicious traffic, where blocking is one step inside your security process, not the whole response. The second is traffic that clearly violates a written policy management actually stands behind. The third is an emergency where nothing else can restore service fast enough. In that case it is a tourniquet, applied loudly, with the fix to follow. Block anything a person uses for work without policy backing and management awareness, and you become the villain of a story. It will be told without your side in it. If a block is genuinely warranted, get the decision made by the person whose name belongs on it. Put it in writing. Make sure the affected user hears it from a human before they discover it from an error message.


ResolutionBest when…Watch out for…Who decides
RescheduleJob is automatic and time-flexibleWindow too short; jobs piling into the same night hoursYou + the job's owner
ThrottleJob must run during shared hoursCaps so tight the job never finishes; forgetting the cap existsYou + the app's owner
RelocateSame bytes crossing the pipe repeatedlySetup effort; a new server to maintainYou, sometimes with budget approval
BlockMalicious, policy-backed, or an emergency tourniquetBreaking work silently; becoming the villainManagement or security — not you alone

When the hog is the business


Sooner or later the workflow points at traffic that is not a misbehaving checkbox but the company doing its job. Think of the design team pushing enormous renders to a client, the media group's footage uploads, or the analytics feed the executives asked for. You do not get to throttle revenue on your own authority, and trying to is a career-shortening move. This is where the conversation moves up a level, to the department head — and where your role changes from enforcer to honest broker of tradeoffs.


Bring the shared-pipe reality in plain language. “We have one pipe; your uploads and everyone's video calls share it; between two and four PM both lose.” Then bring options with costs attached. A transfer window tonight costs nothing. A throttle stretches the job but frees the afternoon. A dedicated or upgraded link costs real money. Let the business weigh business value — that is their job, and they are better at it than you. Yours is making the tradeoffs visible with data. The utilization graphs and per-conversation evidence you gathered are what turn this meeting from opinions into decisions. If the honest answer is that the company has outgrown its pipe, that evidence can start a well-supported upgrade case. Add the guidance in how full is too full, and you have the start of that case rather than a budget argument you lose.


Tip: in that meeting, never present the hog as an offense. Present it as demand. “Your team's work now moves more data than the network was sized for” is a compliment with a budget line attached. Department heads respond to it very differently than to an accusation.


Write it down


Every resolved hog deserves five minutes of writing, in the ticket or your ops log. Not a novel — a record with the facts you will want when something echoes later:


INCIDENT: WAN upload saturated, Tue 09:10–09:55
EVIDENCE: top conversation 10.0.40.23 -> cloud storage, 21 GB
CULPRIT:  MKTG-PC-07 / cloud sync client / new footage folder
VERDICT:  legitimate, badly timed
ACTION:   sync paused, client upload cap set, bulk resumed
          after hours
AGREED:   with K. (marketing) — future footage drops go in
          after 17:00; IT to propose sync policy
FOLLOW-UP: add per-host upload alert on WAN (done Thu)

The payoff arrives on four schedules. Next week, when the same graph shape appears, you diagnose it from memory in one minute. Next quarter, when someone asks “why is the backup throttled?”, the answer exists and has a date and an agreement attached. Next year, when your successor inherits the network, the log is the difference between wisdom and folklore. And continuously: the “agreed” line is quiet armor. Memories of verbal agreements diverge fast. The person who wrote it down at the time wins the divergence politely.


Keep the note wherever your team already searches — the ticket system, the ops log, the runbook wiki. A beautifully written record in a file nobody can find is a record that does not exist.


From one incident to better policy


A hog you fixed is worth something; a hog that upgrades your defaults is worth ten times more. After the ticket closes, ask what rule or safeguard would have made this incident boring, because almost every hog exposes a missing one. A sync client flattened the link at ten in the morning: should client upload caps be part of the standard build? A backup ran at noon: does anything alert when a scheduled job runs outside its window? A department's workload outgrew the pipe silently: should someone glance at trend graphs monthly? The second appearance of the same hog is no longer a user problem — it is a process problem wearing a user's badge.


Feed each answer into the machinery: a default hardened, a threshold alert added, a transfer window published, a line added to onboarding. That pipeline — incident to note to rule — is the entire subject of preventing repeat bandwidth hogs, which turns this article's epilogue into a design discipline.


The conversations that go wrong


Finally, the anti-patterns — each one a real and popular way to convert a solved technical problem into a lasting political one:


  • The ambush. First contact is an email with the user's manager in copy. You have made an enemy and taught a department to route around IT. Talk to the person first; involve managers when policy or repetition genuinely requires it.
  • The public trial. Naming the culprit in the all-staff channel, even jokingly. The next person who suspects their machine is misbehaving will hide it from you.
  • The silent block. Traffic gone, no message, no note. The department's tool “mysteriously broke,” a support case burns a day, and the eventual discovery reads as sabotage. If you must mitigate now, notify as you do it.
  • The jargon dump. Explaining flow exports and interface counters to someone who wanted to know if they are in trouble. Lead with the human meaning: “nothing's wrong — an app on your machine was busier than the network could handle; here's the plan.” Keep the forensics for those who ask.
  • The assumed verdict. Deciding guilt during the walk to the desk. Carry the question, not the sentence; half the time the machine surprises you.

Where to go next


The human step sits between two technical ones. Upstream are the workflow that names the hog and the classification that sets the verdict. Downstream is the prevention design that makes the next incident a routine alert. The whole series lives at Top Talkers: Who's Eating the Bandwidth. Check for five minutes, open with a question, write down what you agreed, and you will step on fewer rakes than I did.


Put this into practice with SysPeep Monitor

Every conversation in this guide goes better with evidence in hand — a graph beats an assertion in any meeting. SysPeep Monitor keeps that evidence flowing with live dashboards and utilization history for every interface it discovers. It also provides reports you can put in front of a department head without translation. The free trial installs on a Windows machine in minutes.

Download Free TrialLearn About SysPeep