An always-on DDoS filter protects your server even when nobody is attacking it. That is also when it can get in the way: it mistakes normal traffic for an attack and drops it. We went through 15,772 attacks from the last 14 months and collected the false positives from our support tickets. This is what we learned.
- Most attacks are short and small: the median attack on a single address lasts 55 seconds and peaks at 1.1 Gbit/s. Nine out of ten are over after about 5 minutes.
- Attacks hit the same few addresses again and again: over 14 months, about 15% of our addresses behind the filter were attacked at least once. 81% of all attacks hit an address that had already been attacked in the 30 days before.
- A filter that learns can be wrong. We have seen new protocols detected as the wrong game, DNS servers that stopped answering and VPN tunnels that kept dropping.
- Not every problem is the filter: customers have run 12 filter tests with Zappy since November 14. The filter was the cause twice, ten times it was not.
- What we do today: filter on by default, can be switched off for services with their own IP, 30 days of mandatory protection after an attack, and a filter test that shows within minutes whether the filter is to blame.
Since September 8, 2025 we store every attack our filtering partner PletX reports: start, end, peak bandwidth and attack type. By November 18, 2026 that added up to 15,772 attacks on 2,712 different addresses.
Left: attack duration in minutes. Right: peak bandwidth in Gbit/s. Share of all attacks on single addresses (chart: ZAP-Hosting)
| Attack type | Share of attacks |
|---|---|
| UDP flood | 77% |
| Amplification (DNS, NTP) | 34% |
| TCP floods (SYN, ACK and similar) | 31% |
| ICMP flood | 26% |
| several types at once | 52% |
Many attacks combine several types, so the shares add up to more than 100%.
What stands out:
- Short does not mean harmless. Thirty seconds are enough to kick every player off a game server. If you only react once an attack has started, you are too late for half of them.
- Bandwidth is rarely the issue. 90% of attacks stayed below 10 Gbit/s. That is little for the filter, but still enough to take down a single server with a 1 Gbit/s port.
- A few addresses attract most attacks. 70% of the attacked addresses were hit exactly once. The top 1% of addresses, on the other hand, took 58% of all attacks.
- Some attacks target whole networks. Five times, the target was not one address but an entire network of 256 addresses at once. We left these carpet attacks out of the numbers above because their bandwidth is spread across many servers.
A filter for game servers has to do more than throw away bandwidth. For every port it learns which protocol runs there, for example FiveM, Minecraft or Valve's Source Engine. From then on it only lets through packets that match that protocol. That is exactly why an attack made of junk traffic never reaches your server.
But learning is an educated guess. When the guess is wrong, the filter treats real traffic as junk. To your players this looks like a broken server: they cannot join, get kicked after a few minutes or cannot find the server in the list. Your logs show nothing, because the packets never reach your server.
All cases come from support tickets of the last few weeks. We measured each of them before blaming the filter.
- A new detection profile, many wrong matches: in October the filter got a profile for QUIC, a protocol Hytale uses among others. In just over 24 hours it learned about 190 ports as QUIC. By our check, real QUIC ran on about 12 of them. Assetto Corsa racing servers were among those affected: their lobby could no longer reach them. PletX tightened the detection within a few days.
- DNS servers that stopped answering: on one VPS the filter learned the DNS port as a game protocol. DNS queries over UDP no longer arrived, while TCP kept working. It took ten days to find the rule as the cause, because nothing was visible on the server itself.
- VPN tunnels that kept dropping: customers running WireGuard, Tailscale or NetBird on their VPS reported tunnels that stopped getting replies after a while. Several cases came down to a learned rule. In others, the test without the filter showed the same problem.
- Hytale connection setup: on some Hytale servers, part of the connection attempts did not complete. Some of that was clearly caused by the learned rule, some of it remained without the filter.
- Websites behind a proxy: a customer with a website behind Cloudflare measured 10 to 20% stalled connection attempts with the filter and none without it. To a filter, lots of connections from a few senders look a lot like an attack.
- A port changes games: on our test systems, different games keep using the same ports. There, the rule learned for the previous game stuck around for a while and blocked the new one.
One more effect surprised us: switching the filter on while players are connected effectively disconnects them. In our measurement, running UDP connections lost every second packet until the client reconnected. So after switching, have everyone reconnect once.
Less often than the tickets suggest. Since November 14, Zappy, our AI support in the chat, can run a filter test: filter off for a moment, check the problem, filter back on. In the first days there were 12 such tests. The problem went away without the filter twice and stayed ten times.
Two measurement mistakes keep coming up:
- One-way UDP tests are misleading. Tools that just send packets to a port and count how many arrive regularly show about 50% loss on our network, even with no filter at all. Only a real connection attempt with a reply from the server tells you something.
- Test packets are not game traffic. On a port with a learned game protocol, arbitrary test packets are dropped on purpose, while real players join without problems. A port checker reporting "closed" does not prove a false positive.
The numbers above suggest an idea: if 85% of addresses were never attacked in 14 months, you could leave the filter off there and switch it on only after the first attack. From then on it would stay on for 30 days, since 81% of attacks hit addresses that were attacked shortly before.
That is exactly what we tried from November 10, first with a few hundred and then with several thousand addresses that had not been attacked in the previous 30 days. During those days there was no attack on any of the opened addresses, and no more support requests than for the filtered ones. We still stopped the test after three days. Every exception is a separate rule on the filter's routers, and several thousand of them pushed the hardware to a limit. The risk for all our other customers was too high.
Then there are the carpet attacks. When an attack hits a whole network unfiltered, it is not just the targeted servers that suffer but everything near them. And 17% of attacks came without any warning: they hit an address for the first time.
- The filter is on by default, for every service and every address.
- You can switch it off yourself if your service runs in our Eygelshoven location and has its own IP, in the DDoS tab of the service. At the end of October, about 640 addresses were on this exception list.
- After an attack, protection is mandatory for 30 days. If an address without the filter gets attacked, we switch the filter on automatically and notify you by email. The panel shows until when protection is mandatory.
- The filter test finds the cause in minutes. You can run it yourself in the panel or let Zappy do it in the support chat, including for game servers on shared addresses. Step-by-step instructions are in our help article Connection issues in Eygelshoven: temporarily disable DDoS protection.
- We report confirmed false positives to PletX, with packet captures and a comparison without the filter. That improves detection for everyone, not just for your server.
- A check script looks for suspicious rules, for example a game server that suddenly has no players after a rule was learned. It runs every 10 minutes and flags anything unusual to our team.
How the filter works in general is explained in our help article PletX DDoS protection at ZAP-Hosting.
- Check whether an attack is running. You can see the status in the DDoS tab of your service.
- Run the filter test: filter off, reproduce the problem, filter back on. Or ask Zappy in the support chat.
- If the problem is gone without the filter, tell our support. We will have the rule corrected.
- If the problem stays, the filter is not the cause. Switch it back on and we will keep looking together.
- Test with real connections, not with port checkers or one-way UDP tests.
How we measured. Attacks: all 15,772 attacks PletX reported on ZAP-Hosting addresses between September 8, 2025 and November 18, 2026, taken from the attack log in our panel. Duration and peak bandwidth refer to the 15,767 attacks on single addresses; the five carpet attacks on whole networks are not included there. Peak bandwidth is the highest value PletX measured per attack. The share of about 15% is based on roughly 17,800 addresses behind the filter as of November 2026; the address pool changed slightly over the period. False positives: cases from our support in October and November 2026, not a complete statistic. We cannot derive an overall rate of how often the filter is wrong from them. Filter tests: all logged filter tests from November 14 to 18, 2026. We only analysed totals, no data of individual customers.
An always-on filter is the right default for game servers, because most attacks are over before anyone could react. It comes at a price, though: it learns, and sometimes it learns wrong. We think that is acceptable as long as you can spot a false positive quickly and it gets fixed quickly. That is what the filter test, the mandatory protection after attacks and our reports to PletX are for.
DDoS protection is included with every ZAP-Hosting game server and VPS at no extra cost.
How long does a DDoS attack on a game server last?
Usually not long. With us, half of all attacks on a single address last under a minute and 89% are over after 5 minutes. Less than 1% last longer than an hour.
What is a false positive in DDoS protection?
The filter mistakes normal traffic for an attack and drops it. Your players then cannot join or get kicked, even though no attack is running.
Can I turn off DDoS protection at ZAP-Hosting?
Yes, for services with their own IP in our Eygelshoven location, in the DDoS tab. After an attack, protection stays on for 30 days.
How can I tell whether the DDoS filter is causing my problem?
With the filter test: turn the filter off for a moment and check whether the problem goes away. If it stays, the cause is somewhere else. Zappy in the support chat can run the test for you.
How strong are DDoS attacks on game servers?
The median attack peaks at about 1 Gbit/s, and 90% stay below 10 Gbit/s. That is still enough to knock a single server offline without a filter.




Discussion
Comments
We'd love to hear your thoughts. Comments appear after a brief review.
No comments yet. Be the first to comment!