DDoS Attack or Traffic Spike? How to Tell the Difference on Your Server
High traffic alone does not prove a DDoS attack. The checks that separate a real flood from a busy server, with the commands to run.
By Frabs · 1 November 2026 · 5 min read
The short answer: look at how the traffic behaves, not how much of it there is. Real visitors complete connections, get answers and come from sources that make sense for your service. A flood tends to leave connections half open, goes unanswered, or arrives from a pattern that normal users would not produce.
Getting this wrong is costly in both directions. Treat a busy day as an attack and you limit your best customers. Treat an attack as a busy day and your server, or the whole node, goes down while you wait.
Why Traffic Volume Is Not Enough
A popular game server, a VPN exit, a large download or a mention on a big website can all produce graphs that look alarming. So can a backup job at 2am. Meanwhile, a small attack aimed at the right place can take a service down with traffic that barely shows on a bandwidth graph.
Volume tells you something is different. It does not tell you what. To find out, you need to look at the shape of the traffic.
How to Spot a SYN Flood
A normal TCP connection completes a three-way handshake: SYN, SYN-ACK, ACK. In a SYN flood, the sender never finishes it, so half-open connections pile up on the server.
Count the connections stuck halfway:
ss -tan state syn-recv | wc -lA handful is normal. Thousands that stay high point to a flood. Then check whether the listen queue is overflowing:
nstat -az TcpExtListenOverflows TcpExtListenDropsA busy website looks different: plenty of new connections, but they complete. You see many in the established state, and data flowing in both directions. To compare, see the overall picture with:
ss -sHow to Spot a UDP Flood or Amplification Attack
UDP has no handshake, so you judge it by what it is and whether anyone asked for it. Capture a short sample:
tcpdump -ni <interface> -c 500 udpTraffic arriving from source ports such as 53 (DNS), 123 (NTP), 1900 (SSDP) or 11211 (memcached), when your server sent no matching queries, is a strong sign of a reflection attack. Those services are being used to send large replies at you.
High UDP on its own proves little. Game servers and VPNs are built on it. What matters is whether the traffic matches the service you actually run on that port.
Signs of a Real Attack and Signs of a Busy Server
Signs of a flood:
- Many half-open connections, and a heavily skewed balance of SYN to ACK packets.
- Replies that never come, with most connections unanswered.
- Very small, uniform packets, or the same request repeated.
- Sources that appear once and never again, which often points to spoofed addresses.
- Traffic that starts suddenly at a level which then holds, with no daily rhythm.
Signs of a busy server:
- Connections complete and data flows both ways.
- Packet sizes and requests vary, as they do with real people.
- The rise matches something you can name: a release, a mention, a campaign or a time of day.
- Sources make sense for your audience, and many of them return.
- CPU and application load rise in line with the traffic you are actually serving.
Also keep in mind that a slow site is not always under attack. A failed release, a full disk or a slow database upstream can look the same from the outside, so check those before you start blocking.
Traffic That Looks Like an Attack but Isn't
These catch out simple threshold alerts all the time:
- Peer-to-peer traffic: many hosts on random high ports can look like a scan, but connections get answered.
- Backscatter: a server under attack sends replies to forged addresses, so a victim can appear to be sending a flood of its own.
- Being scanned: an address on someone else's target list is not an attacker for receiving those packets.
- Shared storage or backups: heavy traffic to one destination, on a schedule, shared by other servers.
What to Do If It Is an Attack
- Confirm the type, using the checks above, so you know what to limit.
- Protect the service. For a SYN flood, make sure SYN cookies are on, and limit new connections per source.
- Limit only the attack. Rate limit the protocol or block the source, rather than cutting off everything aimed at the service.
- Talk to your upstream early if the link itself is full. Nothing on your server can fix a saturated link, so that needs filtering before the traffic reaches you.
- Keep the evidence. Save captures and graphs, because you will want them for the report and for the next time.
To check SYN cookies are on, where 1 means on:
sysctl net.ipv4.tcp_syncookiesHow Frabs Makes This Decision
This is the question Frabs is built around. A threshold only starts an investigation. Frabs then weighs readings for and against an attack, such as whether connections are answered, what the traffic is aimed at and whether it looks like peer-to-peer. If the evidence does not hold up, the finding is left out of your incidents, so a busy day does not turn into an alert or a limit on a good customer.
When it does open an incident, it shows the readings that decided it, what was done and whether that worked. The attack types guide lists what it looks for.
Frequently Asked Questions
How can I tell if my server is under a DDoS attack?
Check how the traffic behaves. Many half-open connections in the SYN-RECV state, unanswered traffic, uniform packets and sources that never return all point to an attack. Traffic that completes connections and matches something you can name is usually real.
How do I check for a SYN flood on Linux?
Run ss -tan state syn-recv | wc -l to count half-open connections. Thousands that stay high suggest a SYN flood, especially if nstat also shows the listen queue overflowing.
Can a small amount of traffic be a DDoS attack?
Yes. An attack aimed at a weak point, such as a login page or a connection limit, can take a service down with traffic that barely shows on a bandwidth graph. Volume alone does not tell you either way.
What causes false DDoS alerts?
Alerts based only on thresholds are often set off by game servers, VPNs, peer-to-peer traffic, backups and real surges in visitors. A server replying to forged addresses during an attack can also look like it is attacking someone else.