Remote-Access VPN Capacity: Licenses, Sessions, and Rush Hour
Every remote-access VPN has a number, and most admins do not know theirs. It is the number of people who can be connected at the same time before the next person gets refused, throttled, or silently broken. The day you learn it the hard way is always a day when the whole company is trying to work from home. The VPN that has hummed along for years at a comfortable load hits its ceiling at nine in the morning during a storm. Suddenly you are rationing connectivity like it is wartime.
The unfair part is that this outage is completely predictable. Capacity problems do not sneak up the way hardware failures do. They announce themselves for weeks in a metric you can graph — concurrent sessions climbing toward a fixed ceiling. You just have to know which ceilings exist (there are three, and the lowest one wins). Watch how close you run to them. Do the small amount of arithmetic that tells you whether an everyone-works-from-home day fits.
This guide walks through exactly that. It covers what "capacity" means for a remote-access VPN, and the three ceilings and how each one fails. It covers the monitoring that gives you weeks of warning. And it covers a planning exercise that answers the only question that matters — are you one snow day away from a bad Monday?
In this guide
- Concurrent sessions: the unit of capacity
- The three ceilings
- Monitoring sessions against the ceiling
- Rush hour: the shape of VPN demand
- An annotated surge morning
- Honest per-user throughput expectations
- Address-pool exhaustion: the sneaky ceiling
- Capacity planning: the snow-day test
Concurrent sessions: the unit of capacity
A quick definition first. A remote-access VPN is the kind individual users connect to. A client on a laptop dials into a gateway (often called a concentrator) at your office or data center. And the user's machine becomes, logically, part of the company network. A remote-access VPN is the one-person cousin of the site-to-site tunnel that joins whole offices together — that kind has its own guide, Monitoring Site-to-Site Tunnels. Each connected user holds a session: an authenticated, encrypted connection that exists from the moment they connect until they disconnect or time out.
The unit of VPN capacity is the concurrent session — how many of those exist at this moment. Not how many employees you have, not how many accounts can log in, not how much bandwidth the circuit has. A thousand users who connect occasionally might peak at eighty concurrent sessions; two hundred users who live on the VPN might peak at two hundred. Capacity questions are always questions about the peak of that concurrent number, because the ceiling does not care about your average. An empty VPN at midnight buys you nothing at nine the next morning.
One subtlety worth catching early: sessions and users are not the same count. One person connected from a laptop and a tablet may hold two sessions; a client that reconnects badly can briefly hold a ghost session alongside its new one. When you compare your monitoring to your intuition ("we only have 150 remote staff, why are there 180 sessions?"), this gap is usually the explanation. It means your planning should be done in sessions, the unit the ceiling is enforced in.
The three ceilings
Your real capacity is the lowest of three separate limits, and they fail in three different styles.
Ceiling one: the license count
Most commercial VPN gateways license concurrent sessions: you bought the right to some number of simultaneous connections, and the gateway enforces it. This ceiling is a hard wall with a clean failure. Session number N connects fine. Session N+1 is refused. Usually it comes with an error message vague enough to send the user to the "my password is broken" theory. The license ceiling is the easiest to hit and the easiest to fix (you buy more), but only if you know it is the one you hit. That is exactly what a graph of sessions flatlining at a suspiciously round number tells you.
Ceiling two: hardware and crypto capacity
Every packet through the VPN is encrypted and decrypted, and that work costs processing power. Gateways have a practical limit on how much encrypted traffic they can push. On bigger boxes that work is handled by dedicated crypto hardware, on smaller ones by the same CPU doing everything else. This ceiling is soft and rude: nothing refuses to connect, but as the box saturates, everyone's connection turns to syrup. Latency climbs, transfers crawl, calls break up, and the ticket queue fills with "the VPN is slow" rather than "the VPN is full." Watch the gateway's CPU and, where exposed, its crypto-engine load alongside session counts. A gateway that pins its processor at the morning peak is at this ceiling no matter how many licenses remain.
Ceiling three: the address pool
Every connecting user must be handed an internal IP address from a pool you configured, and that pool has a fixed size. If the pool holds 200 addresses, user 201 cannot be given one, no matter what the license says. If DHCP scope exhaustion is a familiar idea, this is the same disease in a different organ — and the same slow-motion weirdness applies, as covered in DHCP Scope Exhaustion: Why It Looks Like Bad Wi-Fi. Pools run smaller than you remember, because they were sized years ago for a smaller remote workforce. Disconnected sessions can hold their addresses for a while before returning them, shrinking the effective pool below its nominal size.
Monitoring sessions against the ceiling
The core check is almost embarrassingly simple: read the current session count from the gateway on a schedule, and compare it to the maximum. Most gateways expose the count over SNMP or through an API or status command. SNMP is the standard protocol monitoring tools use to ask devices about themselves. The maximum you find once, in the license details and the pool configuration, and record. Then create two thresholds on the percentage: warn around 70–80% of the ceiling, alert around 90%. The warning is your purchasing signal, weeks in advance; the alert means the next surge probably will not fit.
A live view of a healthy gateway makes the idea concrete — three numbers, one worst-case percentage each:
Sessions: 163 current / 250 licensed (65%) Address pool: 163 in use / 220 addresses (74%) Gateway CPU: 58% (crypto engine 61%) Effective ceiling: 220 (the pool is the lowest wall)
Notice what the last line does: it names the real ceiling, which is the smallest of the three. This gateway's owner might believe they have 250 seats. They have 220, and at the morning peak of 163 they are closer to trouble than the license math suggests. Monitor the percentage against the effective ceiling, and re-derive it whenever a license renews or a pool is resized. And graph the session count over time. That is not because the graph alerts, but because the history is where every capacity answer lives. It holds your true daily peak, your growth trend, and the evidence you attach when you ask for budget.
| Ceiling | How it fails when hit | What to watch | Starting rule |
|---|---|---|---|
| License count | Hard stop: new connections refused | Current vs licensed sessions | Warn 75%, alert 90% |
| Hardware / crypto | Soft: everyone gets slow together | Gateway CPU, crypto load, total throughput | Warn at 70% CPU sustained through the peak |
| Address pool | Connects succeed, then no address or instant drop | Pool addresses in use vs pool size | Warn 80%, alert 90% |
Tip: if you run more than one gateway, monitor each one separately as well as the total. Load rarely balances as evenly as the design promised, and "plenty of capacity overall" is no comfort to the users piled onto the one gateway that is full.
Rush hour: the shape of VPN demand
Once you graph concurrent sessions for a couple of weeks, you will see that VPN demand has a shape. The shape is the whole story. On an ordinary working day it looks like a commuter chart. There is a steep ramp as the workday starts, and a plateau through the morning. Often there is a small dip around lunch. There is a gentle decline through the afternoon, and a long thin tail of evening stragglers. Your capacity problem lives entirely in that morning ramp — the plateau is your true working peak. And it is the number every threshold should be judged against.
Then there are surge days, and they are why headroom exists. Weather that closes roads or schools. A transit strike. A building problem — power work, a burst pipe, an elevator inspection that somehow empties the whole office. A public-health scare. The company all-hands that half the workforce dials into from home. On those days the population that can work remotely becomes the population that does, all at once. On those days, your peak is no longer "the usual remote crowd" but something close to "everyone." Surge days are not anomalies to shrug off. They are rehearsals for each other, and the peak of your worst surge day so far is the most honest capacity requirement you own. Find it in your session history and write it down — it is the number the snow-day test below is built on.
One quiet variable shapes the whole curve: your idle-timeout settings. A gateway that disconnects sessions after a short idle period keeps the concurrent count lean but generates reconnect churn. One that lets idle sessions linger for hours inflates the plateau with people who went to lunch and never came back. Neither choice is wrong, but know which one you made. It decides how honestly your session graph reflects actual working humans. It also decides how much of your precious ceiling is being held by laptops asleep in backpacks.
An annotated surge morning
Here is what a closure-day morning looks like from the monitoring seat, compressed into the entries you would see on the session graph and in your alerts. The gateway in this story has the effective ceiling we computed earlier: 220 addresses, 250 licenses.
07:10 sessions 41 pool 19% normal early crowd
08:05 sessions 118 pool 54% ramp is steeper than usual
08:25 sessions 167 pool 76% WARN: pool above 75%
08:40 sessions 191 pool 87% reconnect churn visible
(drops + rejoins every minute)
09:02 sessions 214 pool 97% ALERT: pool nearly exhausted
09:15 sessions 218 pool 99% new connections now failing
licenses used: 218 of 250
Read the story in it. The ramp at 08:05 was the first clue that this was not a normal day — steeper and earlier than the usual commuter curve. The 08:25 warning bought a real window. For half an hour, an admin who trusted the alert could shorten the address hold time or free stale sessions before anyone was hurt. The churn note at 08:40 is the multiplier at work — flaky home connections dropping and rejoining, each cycle briefly holding two addresses. And the final line is the punchline of the whole guide. The gateway failed with thirty-two licenses to spare, because the pool was the lowest wall. Anyone staring at the license count that morning saw headroom; the pool graph saw the outage coming an hour away.
Honest per-user throughput expectations
Capacity is not only "can they connect" but "is it usable once they do," so let us do the uncomfortable arithmetic. The bandwidth available to VPN users is whatever your internet circuit and gateway can actually push. At peak, that gets divided among everyone connected. A few hundred concurrent users sharing a gateway do not each get a private highway. They get a share. The share at nine in the morning is what users will describe when they say the VPN is slow. Suppose the circuit and gateway together can realistically move a few hundred megabits of encrypted traffic. Suppose two hundred people are connected. If even a quarter of them are actively moving data at the same instant, each active user is living on a small handful of megabits. That is fine for email, documents, and terminal sessions. It is a sad afternoon for someone trying to drag a virtual machine image across the tunnel. Do this division with your own numbers: realistic gateway throughput over peak active users. Look at the result without flinching before you promise anyone anything.
Three honest footnotes belong next to that number. First, encryption overhead: wrapping every packet costs some capacity, so the VPN moves somewhat less data than the raw circuit could. Second, the user's own connection: their home broadband, and especially its upload direction, is often the real bottleneck. It is one you cannot see or fix. The wider problem of visibility into home setups is its own topic, covered in the home-office visibility guide. Third, traffic policy: if every user's video calls and cloud backups ride the tunnel, you are paying for all of it twice over your circuit. Whether that traffic belongs on the VPN at all is the split-tunneling question. The guide Split Tunneling: What You Can and Can't See weighs it properly. Publishing a modest, truthful expectation — "designed for office apps and file access, not for moving huge media files" — saves everyone a winter of disappointed tickets.
Address-pool exhaustion: the sneaky ceiling
Pool exhaustion deserves its own section because its symptoms are the most misleading of the three. The user's connection attempt starts normally. The client reaches the gateway, and the password and second factor are accepted. Then the connection fails at the last step, or connects and immediately drops, or connects with no working access. From the help desk's chair it looks like a client bug or an authentication problem, because "we ran out of addresses" produces no error anyone recognizes. Meanwhile other users — the ones who connected earlier and got addresses — are working fine, which convincingly disproves the theory that the VPN is down.
That pattern is the signature worth memorizing: early birds fine, latecomers broken, auth succeeding, timing dependent. It is a nine-o'clock disease, striking whoever connects after the pool empties. Watch for its accomplice, too: sessions that end without their address returning to the pool promptly. Where the gateway holds addresses for a grace period after disconnect, a morning of flaky home connections can consume the pool fast. In that case, users dropping and reconnecting repeatedly can consume it far faster than the count of actual people suggests. The monitoring is the same pool-usage percentage from the table above. The reason it gets its own alarm is that of the three ceilings it is the one most often sized long ago, forgotten completely, and discovered again at the worst moment. If the warning fires, the fixes are unglamorous: widen the pool or its subnet, shorten the address hold time, or split users across additional pools.
Capacity planning: the snow-day test
Everything above becomes a plan with one exercise. The question: if tomorrow morning everyone who can work remotely does, do they fit? The walkthrough:
- 1. Find your effective ceiling. It is the minimum of licensed sessions, the address pool, and the session count at which the gateway's CPU makes the experience miserable. Your graphs of surge days will suggest that session count — the point where CPU pinned or complaints spiked.
- 2. Find your worst honest peak. The highest concurrent-session number in your history, taken from a real surge day, not an average.
- 3. Estimate the everyone-remote number. Count staff who can plausibly work remotely; that is the demand on a closure day. Sanity-check it against the worst surge you have already seen.
- 4. Compare, with margin. If the everyone-remote estimate is above about 80% of the effective ceiling, you are one snow day from a bad Monday. In that case, the remaining margin is thinner than one bad morning's worth of reconnect storms and forgotten sessions.
- 5. Close the gap on a calendar, not during an outage. More licenses, a bigger pool, a second gateway, or a policy change that lightens each session's load. Each has lead time; none is available at nine in the morning in the snow.
Run the test twice a year and after any growth spurt, and keep the arithmetic in your documentation next to the session graph. The pleasant surprise is how cheap foresight is here: one SNMP counter, two thresholds, and ten minutes of arithmetic replace the entire genre of surprise capacity outage.
Where to go next
Capacity is the second question of remote-access monitoring. The first is whether the service is healthy at all. The day something breaks you will want a triage order rather than a guess — that is A VPN Troubleshooting Playbook: From "VPN Is Down" to Root Cause. The rest of the pillar — tunnel truth-telling, per-tunnel throughput, and the split-tunneling visibility trade — lives at VPN & Remote-Access Monitoring. Watch one number against three ceilings, respect the morning ramp, and let the snow day find you bored instead of besieged.
Put this into practice with SysPeep Monitor
Session counts, pool usage, and gateway CPU are exactly the kind of numbers SysPeep Monitor collects all day over SNMP and similar protocols. Set headroom thresholds once and it warns you weeks before the ceiling. Its dashboards keep the session graph — and your capacity evidence — one click away. The free trial runs on any Windows machine.
Download Free TrialLearn About SysPeep