Split Tunneling: What You Can and Can't See
A remote user reports: "the internet is slow when I'm on the VPN." Before you can even begin to troubleshoot that sentence, you need an answer. The question is one many admins have never written down: when this person is connected, does their internet traffic actually pass through you? Depending on one policy setting on your VPN, the honest answer could be "yes, all of it, and it is our problem." Or it could be "no, none of it, and we physically cannot see what they are experiencing." Same complaint, two different realities.
That setting is split tunneling. It is one of the most consequential choices in remote access — not just for security or bandwidth, but for what your monitoring can and cannot observe. It decides which complaints are yours, which graphs mean anything, and whether you can prove your innocence when remote users have a bad day. Yet it is usually discussed as a security topic and almost never as a visibility topic.
This guide fixes that. It explains what full and split tunneling are in plain language. It covers the visibility you gain and lose with each and what to monitor in each mode. It shows how to verify the policy is actually doing what the document says. And it explains how to help users troubleshoot the half of their world you cannot see. It ends with an honest decision table from the operations chair.
In this guide
- Full tunnel vs split tunnel, in plain language
- Why organizations pick each one
- The visibility trade
- Monitoring a full-tunnel deployment
- Monitoring a split-tunnel deployment
- Verifying the policy is actually applied
- Helping users with the half you can't see
- The honest decision table
Full tunnel vs split tunnel, in plain language
When a remote user connects to your VPN, their computer has two possible roads for every packet. One is through the encrypted tunnel to your network. The other is straight out their home internet connection like any normal traffic. The tunneling mode is simply the rule that decides which traffic takes which road.
Full tunnel means everything goes through you. Every packet the laptop sends is encrypted, carried to your VPN gateway, and released onto the internet from your network. That includes the connection to your file server, but also the video call, the personal email, the music stream. The user browses the web from your public address, through your circuit, both directions, all day. Split tunnel means only work traffic comes to you. The VPN client is given a list of company networks (and sometimes specific applications). It sends only traffic for those destinations through the tunnel. It lets everything else go directly out the user's own internet connection.
The commuting analogy holds up well. Full tunnel is a rule that every errand in your life must route through the office first. You drive in, then go to the dentist from there, and return via the office on the way back. Split tunnel is normal life: you go to the office for office things and everywhere else directly. The office-first rule gives the office perfect knowledge of your travels and total congestion in its parking garage. Normal life is efficient and mostly invisible to your employer. Every tradeoff below is that comparison wearing different clothes.
Real deployments come in shades. Some split by inclusion: a list of company networks goes through the tunnel, everything else goes direct — the classic form. Some split by exclusion: everything goes through the tunnel except a list of heavyweights, typically video conferencing and other real-time cloud services that suffer most from the detour. Some clients can even split per application rather than per destination. For this guide the shades matter less than the principle. For any given flow, either it crosses your infrastructure or it does not. Everything below follows from which side of that line each kind of traffic lands on. When you read "split tunnel" here, read it as "some meaningful traffic bypasses you," in whatever pattern your policy draws.
Why organizations pick each one
Full tunnel gets chosen for control. All traffic passes the organization's firewall and filtering. Egress comes from known addresses that partner systems can allowlist. Policy is uniform whether a laptop sits in the office or a kitchen. Some organizations are required to run this way by contract or regulation. The costs are physical. Every remote user's whole internet life rides your circuit and gateway twice. That is in from their home, back out to the internet, and the reverse for replies. And every cloud app takes a detour through you. That "hairpin" adds latency the user experiences as the VPN making everything slower, because it literally does.
Split tunnel gets chosen for efficiency. Your gateway and circuit carry only work traffic. Video calls to cloud services take their short natural path and perform better. The VPN's capacity ceiling stretches much further for the same hardware. That is the sessions-and-throughput arithmetic in Remote-Access VPN Capacity. The cost is the loss of the control and uniformity above, plus the visibility consequences this guide is really about. Both choices are legitimate; what is not legitimate is making the choice and then monitoring as if you had made the other one.
The visibility trade
Here is the part that rarely makes the policy document. With split tunneling, the user's internet problems are invisible to you. Their video call stutters, their cloud storage crawls, the news site will not load. None of that traffic ever touches your tunnel, your gateway, your circuit, or your graphs. You could have perfect monitoring on everything you own and still see nothing, because nothing of theirs passes through anything of yours. The complaint arrives addressed to "the VPN" anyway, because the VPN icon is lit and users reasonably blame the most technical-looking thing on screen.
The flip side is genuinely valuable: those problems are also provably not yours. Suppose the tunneled half is healthy: your through-tunnel probes clean, your gateway relaxed, your circuit quiet. If the complaint concerns traffic that never routes through you, your monitoring data clears your network with unusual crispness. This is the exoneration pattern: positive evidence of your own health, assembled before anyone asked. That evidence is exactly the kind The Exoneration File teaches you to keep. Split tunneling narrows what you can see, but it also narrows what you can be blamed for — provided you can show the narrow part was clean.
Full tunnel inverts the trade completely: you can see everything, and you own everything. The user's slow afternoon of web browsing is now legitimately a question about your egress filtering, your proxy, your circuit's afternoon load. Visibility and responsibility arrive welded together; neither mode gives you one without the other.
Monitoring a full-tunnel deployment
If you run full tunnel, size your monitoring to the fact that you have become every remote user's ISP. Watch the capacity chain hard: concurrent sessions, gateway CPU and crypto load, and above all the internet circuit in both directions. Remote users' downloads are your downloads now, twice. Watch the hairpin path. A synthetic test that reaches a representative cloud service through the VPN, timed, tells you what remote users actually experience when they open the tools they live in. The test's history distinguishes "the VPN added its usual small delay" from "the detour got expensive this week." And watch egress health. Whatever filtering or proxying the traffic passes on the way out is now part of every remote user's every click. So its load and errors belong on the same dashboard.
Expect your traffic graphs to carry the whole texture of human internet use — lunchtime spikes, meeting-hour video surges. Set thresholds accordingly. The upside: when a remote user says "the internet is slow," you have real graphs of the actual path their packets took. In full tunnel, that sentence is always at least partly your business, and your monitoring can always speak to it.
Monitoring a split-tunnel deployment
Split tunnel monitoring is leaner and more focused. It covers sessions and gateway health, plus through-tunnel probes to the internal applications users actually reach over the VPN. It also covers the (much lighter) load those flows put on your circuit. What changes most is triage. Every remote-user complaint now begins with one sorting question: does the thing that is slow travel through the tunnel? Internal apps, file shares, anything on company address space — yours; investigate with your graphs. Cloud and internet things — not on your path; move to helping the user examine their side (more on that below). Teach the help desk this question and half of all remote tickets sort themselves in the first minute.
One habit keeps the sorting honest: publish, in the runbook, a current list of what is tunneled — the destination networks and, if used, the applications. Complaint triage, policy verification, and user communication all hang off that list, and a stale list quietly breaks all three.
Give the tunneled half proper synthetic coverage, because it is now the entirety of what you can vouch for. A probe per key internal application, run through the VPN path, gives you a truthful answer to "is remote access working right now." That answer does not depend on waiting for complaints. The probe checks not just "can the gateway be reached" but "does the file share answer, does the intranet respond, and how fast." If your monitoring platform supports remote probes, one placed outside the office network — connecting in the way a user would — is the gold-standard vantage point. That probe exercises the whole chain, client to gateway to application. Its history is the baseline you compare every anecdote against.
Know the DNS wrinkle, too, because it generates the strangest tickets in split mode. Name lookups have to split just like traffic. Queries for company names should go through the tunnel to your internal DNS (otherwise internal apps are unfindable). Everything else resolves via the user's home setup. Clients handle this with rules about which name domains use which resolver. When those rules misfire the symptoms are wonderfully confusing — the tunnel is perfect, routes are perfect, and yet "nothing works" because nothing can be found. When a split-tunnel user can ping internal addresses but not open anything by name, the tunnel is innocent; look at the name-resolution split first.
Verifying the policy is actually applied
The policy document says "split, with these networks tunneled." Is that what laptops are actually doing? Drift here is common and quiet. An updated client ships different defaults, or a config push misses a group. Someone edits the tunneled-network list and fat-fingers a subnet. Or a troubleshooting session leaves one machine switched to full tunnel forever. Each divergence changes both security posture and your monitoring picture, so verification belongs on a schedule, not just at rollout.
The direct check is the client's routing table — the map every computer keeps of which road each destination takes. On a connected laptop, list the routes and read the story. Company networks should point at the VPN adapter. The default route (the "everything else" entry, written 0.0.0.0/0) should point at the home network in split mode — or at the VPN in full mode. Conceptually you are looking at this:
Destination Via Meaning 0.0.0.0/0 home router everything else: straight to the internet 10.0.0.0/8 VPN adapter company networks: through the tunnel 172.16.40.0/24 VPN adapter the app subnet: through the tunnel
Two behavioral tests confirm what the table claims. A traceroute (the tool that lists the hops a packet takes) to an internal server should show the tunnel. A traceroute to a public site should show the home router first in split mode. And any "what address does the outside world see me as" lookup settles it instantly. Your public address means that traffic is hairpinning through you; the user's home address means it is going direct. Have the help desk capture all three from one or two volunteer machines after every client update. It takes five minutes and catches drift while it is still an anecdote.
From the gateway side, your own graphs verify policy in aggregate. Per-session and total throughput have a characteristic size in each mode. If a client update accidentally flips a crowd to full tunnel, your gateway traffic and session throughput jump for no business reason. In that case, a threshold on "traffic per session far above baseline" turns a silent policy change into a Tuesday-morning question instead of a month of mysterious congestion.
Tip: when someone toggles a machine to full tunnel to debug something, make "switch it back" part of the ticket's closing checklist. Every long-lived full-tunnel straggler is a small standing tax on your gateway and one more machine whose behavior does not match the documentation.
Helping users with the half you can't see
In split mode, a real share of remote complaints concern the invisible half, and "not my traffic" is both true and useless as a reply. What works is giving users a short self-test you scripted in advance. Have them try the slow thing with the VPN disconnected: still slow means their internet or the far service, not the tunnel. Have them try a second device on the same task: both slow points at the home network or ISP; one slow points at the machine. If they can, plug into the router by cable: problem vanishes, it was their Wi-Fi. Each test is something a non-technical person can do in two minutes with the help desk on the phone. And each one cleanly moves the investigation without anyone needing access to a home network you will never see. The broader discipline of supporting workers whose networks you do not control — expectations, tooling, and where the visibility boundary really sits — is its own topic. It is covered in the home-office visibility guide.
Meanwhile, your contribution is the clean half: probe results and gateway graphs from the same time window, showing the tunneled path healthy. Pair the user's self-test with your evidence and most "VPN is slow" tickets resolve into something specific. Usually it is the home Wi-Fi, sometimes the ISP, occasionally a genuinely sick service. That happens without a single accusation bouncing off your network on the way.
A worked example, because the choreography matters more than the theory. Ticket: "video calls break up whenever I'm on the VPN." First, the sorting question against the runbook list: video conferencing is excluded from the tunnel, so those packets never touch you. Second, your evidence: through-tunnel probes to internal apps flat and clean all morning. Third, the user's two-minute test: calls break up identically with the VPN disconnected, and also on their tablet. Verdict, reached in one phone call with zero finger-pointing: the home Wi-Fi or ISP, with the VPN present only as a coincidence of timing. It was connected while they worked, so it collected the blame. Write that flow down once and every desk technician can run it.
The honest decision table
Security and compliance get a vote elsewhere; this table is the operations and monitoring view, mode by mode, with no winner declared — because there is not one.
| From the ops chair | Full tunnel | Split tunnel |
|---|---|---|
| User's internet problems | Visible to you — and yours to fix | Invisible to you — and provably not yours |
| Gateway and circuit load | Heavy: all traffic, both ways, twice | Light: work traffic only |
| Capacity headroom per box | Consumed quickly; plan hard | Stretches much further |
| Cloud app experience | Hairpin latency; monitor the detour | Native path; not on your graphs |
| Complaint triage | Everything is plausibly yours | One sorting question splits tickets fast |
| Exoneration when slow | Harder: you are always on the path | Clean: your half has its own evidence |
| Policy verification burden | Low: everything tunnels | Real: the tunneled list must stay true |
If the organization asks the network team which mode to run, answer with this table plus your capacity numbers. In that case, insist only on one thing: whichever mode wins, the monitoring must be built for that mode, and the help desk must know which mode it is. Most of the pain attributed to split tunneling is really the pain of running one mode while troubleshooting as if it were the other.
Where to go next
Split tunneling decides what your monitoring can see. The sibling guides cover how to watch what remains squarely yours. They cover capacity and rush hour for the load side. They also offer the VPN troubleshooting playbook for the day the tunneled half itself misbehaves. The full set of guides lives at VPN & Remote-Access Monitoring. Know your mode, monitor for that mode, keep the tunneled list true, and hand users a flashlight for the half of the path that belongs to them.
Put this into practice with SysPeep Monitor
Whichever mode you run, your half of the path deserves real monitoring. SysPeep Monitor watches the gateway, the circuit, and synthetic probes through the tunnel. Its dashboards give you the timestamped evidence for the is-it-us conversation. It installs on one Windows machine and discovers your network. The trial is free.
Download Free TrialLearn About SysPeep