VPN vs. Direct: Monitoring Both Paths Remote Workers Take
Picture the ticket: "the VPN is slow but my internet is fine." The user is right about both halves, and I spent my first year in remote support treating it as a contradiction. It is not. A remote worker's laptop has two roads out of the house. Traffic for internal systems goes through the VPN tunnel to your office. Traffic for cloud applications usually goes straight out through the home ISP to the vendor. The two roads share a driveway and nothing else. So "my connection is slow" can mean one road, the other, or both. The evidence for each lives in a different place.
Most of the confusion in remote support comes from checking the wrong road. This guide lays out both and what to watch on each. It covers the four things to monitor on the VPN path and the three on the direct path. It explains where each measurement has to be taken from. It also shows how to read the two classic complaints, the one above and its mirror image. You will finish with a map of your applications to their roads. That way, when a user names an application, the help desk knows which road to check before asking a question.
In this guide
- The two roads out of the house
- What to monitor on the VPN road
- What to monitor on the direct road
- The classic case: VPN slow, internet fine
- The reverse case: internet slow, VPN fine
- Mapping applications to paths
- The path-by-check table
- Where to go next
The two roads out of the house
The VPN road starts at the VPN client on the laptop. It crosses the home network and the home ISP inside an encrypted tunnel. It arrives at your VPN concentrator (the device at your edge that terminates tunnels). From there, traffic behaves like any office desktop: file shares, internal web applications, the print server, internal DNS names. Every tunnel from every remote worker shares the concentrator and the office internet circuit it sits behind. That makes those two things the choke points of the whole road.
The direct road starts at the same laptop. It crosses the same home network and ISP, and goes across the internet to a SaaS provider's front door without touching your office at all. That covers cloud email, document collaboration, video meetings, the CRM. Nothing on this road belongs to you. (You will be asked about it anyway.) Which road a given destination takes is decided by the VPN client's routing policy. In a split tunnel, only internal destinations enter the tunnel and everything else goes direct. In a full tunnel, everything enters the tunnel and SaaS traffic makes a detour through your office before heading back out. That policy decides what you can and cannot see. The guide on split-tunnel visibility covers the trade in depth, and here we take the policy as given.
You can see which road a destination takes from the laptop itself. On Windows, ask the routing table which interface would carry traffic to an address:
PS C:\> (Find-NetRoute -RemoteIPAddress 10.0.4.20)[1].InterfaceAlias Corporate VPN PS C:\> (Find-NetRoute -RemoteIPAddress 203.0.113.10)[1].InterfaceAlias Wi-Fi
The internal file server goes out through the tunnel interface; the cloud application's address goes out through the Wi-Fi adapter directly. In a full-tunnel setup the second answer would also say the VPN, and that single difference explains a whole category of tickets.
What to monitor on the VPN road
The VPN road has one great advantage: half of it is yours. You cannot see the home network from the concentrator, but you can see every tunnel arrive, and you own everything after that. Four checks cover it.
Tunnel state, from both ends
From the endpoint: whether the VPN client reports a connected tunnel, sampled every few minutes by the lightweight endpoint check. From the concentrator: the session table, which lists every connected user, when they connected, and from which public address, plus the log of disconnects and reconnects. A tunnel that reconnects every twenty minutes is invisible to the user, who just notices that things stall, and obvious in the concentrator log. The client's status light stays green the whole time, being an optimist. Count reconnects per user per day; a rising count for one person is their home network or ISP. And a rising count for everyone is your edge.
Concentrator capacity
A concentrator has three ceilings. One is a license count for concurrent sessions. Another is a hardware limit on how much encrypted traffic it can push. The third is an address pool to hand out to tunnels. Monitor concurrent sessions against all three, plus the device's CPU, over SNMP. Graph them across the working day so that the morning peak is a known shape rather than a surprise. Finding each ceiling and planning for the snow day when everyone connects at once is the subject of remote-access VPN capacity.
Per-user round trip through the tunnel
From the endpoint, a few pings through the tunnel to an internal host that exists only inside, such as a domain controller. This is the user's actual experience of the road: home network, ISP, internet, concentrator, and office LAN in one number. Subtract the gateway round trip and the remainder is the path plus your edge. Some concentrators also report per-session round trip or retransmissions from their side. Where they do, that number is the same road measured from the other end. In that case, the two should agree.
Internal application response from the endpoint
Time one small request to an internal web application through the tunnel, once per interval. If the tunnel round trip is normal and this number is slow, the application or its server is the problem and the network is exonerated. Pair it with the office internet circuit's utilization, polled over SNMP from the edge router or firewall. Every tunnel and every hairpinned SaaS session shares that one pipe. The pipe carries the tunnels, the meetings, and, on the day it fills, the blame.
What to monitor on the direct road
The direct road is harder, because none of it is yours and the concentrator never sees it. The vantage point has to be the laptop, with a second opinion from your office.
SaaS reachability from the endpoint
A small scheduled request to each critical cloud application's login page, timed, from the laptop. Not a full login, just "does the front door answer and how long does it take." Do this for the handful of applications the business actually depends on. Stagger the requests so a few hundred laptops do not knock on the same door in the same second. The measurement is the home network plus the ISP plus the internet plus the vendor. And the gateway round trip tells you how much of it is the house.
DNS, and which resolver answered
DNS on a remote laptop is a moving target. With the tunnel down, names resolve through the home router and the ISP's resolver. With the tunnel up, the VPN client usually points the laptop at your internal resolver, or at least sends internal names there. Get that wrong and internal names fail while the tunnel looks healthy, or every public name takes a detour through the office. I once spent an hour on a "VPN outage" that was a laptop asking the kitchen router for an internal name. Record which server answered and how long it took, and alert when the resolver changes unexpectedly while the tunnel state does not.
The same check from your office
A synthetic check against the same SaaS login pages from a probe in your office gives you a baseline that excludes every home network. When the endpoint number is bad and the office number is fine, the trouble is on the user's side of the internet. When both are bad, the vendor or the internet near them is the suspect. In that case, the us-or-them playbook takes it from there.
Tip: in a full-tunnel deployment the "direct" road does not exist for the user, but it still exists for you. The SaaS traffic leaves through your office circuit. Monitor that circuit's utilization as a VPN metric. When it fills up, every cloud application slows down for every remote worker at once. In that situation, none of them will describe it as a VPN problem.
The classic case: VPN slow, internet fine
This is the most common remote-work complaint, and it is usually accurate. The user's direct road is healthy: web pages load, cloud email is snappy, video streams. The VPN road is not: file shares crawl, the internal application spins. Since the two roads share the house and the ISP, the difference has to be somewhere they diverge, and that is at your edge. Four causes account for nearly all of it.
- Concentrator load. Sessions near the license ceiling or the CPU pinned by encryption work. Every tunnel slows a little, and the slowing is worst at the morning peak. Evidence: the session and CPU graphs, and tunnel round trips rising for everyone at once.
- A saturated office circuit. Every tunnel shares the office internet link, and in full-tunnel setups so does every cloud session. When that circuit fills, tunnel round trips climb and loss appears for all remote users while their gateway numbers stay perfect. Evidence: circuit utilization at its ceiling on the SNMP graph, lined up in time with the complaints.
- Hairpinned traffic. In a full tunnel, a video meeting travels from the house into the tunnel, through your office, out to the vendor, and back the same way. That puts the meeting on your circuit in both directions and adds a detour through the office to every packet. Evidence: cloud applications are slow only while the tunnel is up, and the office circuit is busiest during meetings.
- The internal application itself. The tunnel is fine; the server behind it is not. Evidence: tunnel round trip normal, application response slow, and office users noticing too.
Complaints arrive from a dozen remote users at ten in the morning. Their endpoint rows show gateway round trips of a few milliseconds with no loss, so the houses are fine. Tunnel round trips that were thirty milliseconds at eight are now well over a hundred with occasional loss, for every user regardless of ISP. So the common element is your edge. The concentrator's session count is high but under its license; its CPU is comfortable. The office circuit's outbound utilization has been flat against its ceiling since half past nine. That is the answer: the pipe is full. The cause turns out to be the full-tunnel policy dragging every morning's video meetings through the office. The fix is to route the meeting application directly, or to buy a bigger circuit. And the graph that showed the ceiling is the business case for either.
The reverse case: internet slow, VPN fine
Less common but more puzzling: the internal application is fine through the tunnel, while cloud applications and ordinary web pages crawl. Here the roads diverge at the house, not at your edge. The tunnel carries its traffic to a destination with a good internet connection. The direct road depends entirely on the home ISP's own routing and peering, which is to say on people who have never heard of your CRM.
- Home ISP congestion. Consumer lines slow down in the evening and on rainy weekends when the neighborhood is streaming. The tunnel suffers too, but a small internal application tolerates it better than a video meeting does. Evidence: throughput samples and SaaS response times worsen at fixed hours; gateway numbers stay clean.
- Poor peering on the direct path. The home ISP's route to a particular vendor is bad while its route to your office is fine. Evidence: one cloud application is slow for users on one provider, and fine for everyone else; a path test from the laptop shows the jump beyond the ISP.
- The home resolver. Public names resolve through a slow or flaky ISP resolver while internal names go through your resolver over the tunnel. Evidence: DNS timings differ by resolver in the endpoint rows.
- A hotspot or hotel. The public IP changed. Everything direct is slower and rate-limited; the tunnel hides some of it by carrying the office traffic to a better-connected place.
A short example: one user on a rural provider reports that the CRM has been unusable for a week, while the internal ticketing system through the tunnel is fine. Her endpoint rows show a clean gateway, a normal tunnel round trip, and a CRM login page taking several seconds from the laptop. The office probe reaches the same page in a fraction of a second. A path test from her laptop shows the delay beginning two hops past her ISP's network. The verdict is the direct path, the owner is nobody you can call, and the workaround is the one below.
This case has an ironic fix. If the office's internet connection is genuinely better connected than the user's ISP, sending that one application through the tunnel can make it faster, at the cost of your circuit. Whether that trade is worth it depends on how many users share the problem and how much headroom your circuit has, which is exactly why you monitor both.
Mapping applications to paths
The help desk does not think in roads; it thinks in the names users say. So build a map from application to road and put it where the first-line technician can see it. The map comes from two sources: the VPN client's routing policy, which lists what enters the tunnel, and your application inventory, which lists what people use. For each application, write down its road, the first check for that road, and who owns the fix. Build it before you need it. The day you need it is a bad day for cartography.
| What the user names | Road | First check | Second check | If bad, owner |
|---|---|---|---|---|
| File shares, mapped drives | VPN | Tunnel state and tunnel round trip from the endpoint | Concentrator sessions and CPU; office circuit | You |
| Internal web application | VPN | Tunnel round trip vs application response | Application server health | You (server team) |
| Internal names not resolving | VPN (DNS) | Which resolver answered | VPN client DNS configuration | You |
| Office printing | VPN | Tunnel state | Print server reachability | You |
| Cloud email and documents | Direct (or VPN if full tunnel) | SaaS reachability from the endpoint | Same check from the office probe | The vendor, the ISP, or the house |
| Video meetings | Direct, ideally | Gateway round trip and loss | Confirm the meeting traffic is not hairpinned | The house, or you if it hairpins |
| The CRM or line-of-business SaaS | Direct | SaaS reachability from the endpoint | Office probe; DNS timing | The vendor or the ISP |
| "Everything" | Both | Gateway round trip and loss; public IP | Hotspot test | The house or the ISP |
To build the "road" column, connect a laptop to the VPN and print its routing table. The routes the client injected are the destinations that enter the tunnel, and everything else goes direct. Do it once per client configuration, not once per user, and repeat it after every policy change. If the injected routes include a default route, you are in full tunnel and every row in the table reads "VPN."
Two rows deserve a note. Video meetings belong on the direct road in almost every design, because real-time traffic hates the extra round trip and the shared circuit. If your map says "VPN" for meetings, that is the first thing to change. And "everything" is a road of its own. When both roads are slow at once, the shared segments are the suspect. In that case, the guide on home Wi-Fi versus corporate blame is the one to open.
At Harbourview Clinic the remote billing team's video calls froze every morning for a month, and every ticket was closed as "home Wi-Fi," because nobody had built the map. When someone printed the routing table, the meeting service was on the VPN road, hairpinning through an office circuit that was flat against its ceiling by nine. The policy change was three lines long. The next morning the team's calls were clear. The circuit graph dropped by half. And the review noted that the home Wi-Fi had "recovered unusually quickly."
Tip: the map goes stale the moment someone changes the VPN client's split-tunnel policy or moves an application to the cloud. Put a line in the change process: any change to routing policy or application hosting updates the map on the same day. A map the help desk cannot trust is worse than none, because it sends them confidently down the wrong road.
The path-by-check table
The checks themselves, grouped by road, with where each runs and what a bad result points to: the monitoring configuration for the remote program on one page.
| Road | Check | Runs on | Method | Bad result points to |
|---|---|---|---|---|
| VPN | Tunnel state | Endpoint and concentrator | Client status; session table and disconnect log | Home network or ISP (one user); your edge (many users) |
| VPN | Concurrent sessions, CPU, address pool | Concentrator | SNMP polling, graphed against ceilings | Capacity: license, hardware, or pool |
| VPN | Tunnel round trip and loss | Endpoint | Pings through the tunnel to an internal host | Path or edge, once gateway is subtracted |
| VPN | Internal application response | Endpoint | Timed request to an internal web page | The application, if tunnel round trip is normal |
| VPN | Office circuit utilization | Edge router or firewall | SNMP interface counters | Saturation: every remote user at once |
| Direct | SaaS reachability | Endpoint | Timed request to each critical login page | House, ISP, internet, or vendor |
| Direct | SaaS reachability | Office probe | Same request from your network | Vendor or internet, if bad here too |
| Direct | DNS time and resolver | Endpoint | Timed lookup, recording the answering server | Home resolver, or VPN DNS misconfiguration |
| Both | Gateway round trip and loss; public IP | Endpoint | Pings to the first hop; echo endpoint | The house; or a hotspot, hotel, or travel |
Half of these rows run on the laptop and half on equipment you already own. The endpoint sees the roads from the driver's seat; your infrastructure sees the traffic arriving. And the two together let you say which road, which segment, and whose problem, usually before the user finishes describing the symptom.
Where to go next
Two roads, four checks on the one you half own, three on the one you do not, and a map from application names to roads. With that in place, "the VPN is slow but my internet is fine" stops being a contradiction and becomes a ten-second lookup. The road model sits inside the larger visibility model. The endpoint checks it relies on are set up in Lightweight Endpoint Checks for Remote Laptops. And the support playbook turns the application map into intake questions. The whole series is collected at Monitoring Remote & Home-Office Workers.
Put this into practice with SysPeep Monitor
The VPN road runs through equipment you own. That half of the story is exactly what SysPeep Monitor is built to watch. It discovers your concentrator, firewall, and edge router. It polls sessions, CPU, and circuit utilization over SNMP, and graphs them on live dashboards. It alerts when the office pipe fills or the session count nears its ceiling. All of this runs from one Windows machine with a single web console. The free trial is a quick download.
Download Free TrialLearn About SysPeep