“DDoS protected” on a hosting plan can mean four very different things: a firewall rule that drops obviously malformed packets, a per-IP rate limiter, a cloud scrubbing network with a few hundred Gbps of capacity, or a blanket always-on transit clean pipe. The differences only become visible during an attack, which is the worst time to discover you bought the first category while expecting the third. This article breaks down the actual mitigation mechanisms, how scrubbing centres work, and what to ask a provider before signing.
The Four Mitigation Layers, In Order of Depth
Protection is layered, and each layer catches attacks the layer above it misses. Providers differ mostly in how many layers they operate and where the layers sit relative to your server.
- Layer 1 — edge filtering at the host. ACLs and stateless filters on the provider’s border routers. Cheap, catches malformed packets and obvious amplification floods, does nothing for a well-formed 200 Gbps SYN flood.
- Layer 2 — per-IP rate limiting and connection tracking limits. Drop or throttle above a threshold. Effective against small-to-medium floods; can be tuned to hurt legitimate game traffic if applied crudely.
- Layer 3 — scrubbing centre diversion. Traffic is re-routed via BGP to a cleaning facility during an attack, filtered, then tunnelled or routed back clean. This is what most people mean by “DDoS protection.”
- Layer 4 — always-on inline filtering. All traffic passes through a filtering device permanently. Lower latency variance than on-demand diversion, but the filter’s throughput becomes your ceiling and a single point of failure.
On-Demand vs Always-On Diversion
On-demand diversion only activates when the provider’s detection system sees an anomaly. Detection typically takes 3–15 seconds from the start of the attack; during that window, traffic reaches your server unfiltered. For a Volumetric flood big enough to saturate your upstream, that window means your link is completely saturated for up to 15 seconds, which for a game server is already a disconnect event for everyone.
Always-on filters trade a permanent small latency cost (usually 1–5 ms depending on scrubbing centre proximity) and a hard throughput ceiling for the absence of that detection window. For a game server in a contest with a persistent attacker, always-on is worth the millisecond; for a web application, on-demand is usually fine.
| Property | On-demand scrubbing | Always-on scrubbing |
|---|---|---|
| Time exposed before filtering | 3–15 s | 0 s |
| Added latency | 0 ms when idle | 1–5 ms always |
| Capacity ceiling | Scrubbing pool capacity | Filter device capacity |
| Failure mode | Detection miss | Filter outage = total outage |
| Good for | Web, API, mail | Game servers, real-time UDP |
How a Scrubbing Centre Actually Works
Your server announces its IP block via BGP. The provider’s scrubbing centre is a peer on that path. When an attack is detected, the provider either withdraws your normal announcement and re-announces it from the scrubbing centre, or splits your /32 so that the first N packets are sampled before the decision. Clean traffic is then forwarded to you either by GRE tunnel or by direct routing, depending on the provider.
Two consequences follow from that architecture. First, all traffic during an attack flows through the scrubbing centre, so capacity is limited by the centre’s link, not by the total of your server’s 10G port. Second, GRE-tunnel backhaul adds roughly 2–8 ms of latency on average, and can add jitter if the tunnel is not well provisioned.
- Anycast scrubbing — multiple sites announce the same /24; traffic lands at the nearest, which gives you an automatic capacity multiplier and low latency.
- Unicast scrubbing — one site, usually in a single region. Cheaper, but your whole attack must be absorbed there.
- Scrubbing-to-source — some providers route each attack’s cleaning site based on the attack’s source region. Useful against globally distributed botnets.
Layer 7 Protection Is a Different Product
Volumetric filtering does nothing for an application-layer attack, because the requests are perfectly valid TCP connections that simply arrive at 40,000 per second. Layer 7 defence needs something that understands the protocol: a generic HTTP WAF is useless against a Minecraft handshake flood, and a Palworld server needs a filter that can distinguish a real join from a spoofed one. Ask explicitly whether the provider can filter at the application layer for your game’s protocol, and how many times they have written such a rule.
The Numbers That Actually Matter in a Quote
| Metric | What “good” looks like | What to watch for |
|---|---|---|
| Declared scrubbing capacity | 500 Gbps – 2 Tbps | Capacity shared across all customers; peak-time contention |
| Detection time | < 5 s | Thresholds tuned to ignore small attacks < 2 Gbps |
| Clean-traffic latency add | 1–5 ms | GRE backhaul through an unrelated region |
| Contractual cover | Unlimited, written into SLA | “Up to X Gbps” with overage billing |
| Protocol filters | L3/4 + L7 for your game | “L3/4 only” buried in fine print |
| Alerting | Email + API + webhook | Silent mitigation; you learn afterwards |
What a Volumetric Attack Does to an Unprotected Box
A 10 Gbps flood against a server on a 1 Gbps uplink does not just slow it down; the provider’s switch may disable the port entirely to protect the access layer. That produces a “null route” — your server is unreachable, sometimes for hours, while the provider waits for the attack to stop. If the mitigation strategy is “null route the customer,” you have no protection at all, only a cost-transfer mechanism. Ask whether null-routing is the fallback when mitigation capacity is exceeded.
Comparing Provider Tiers
Pricing for protected dedicated servers clusters into three bands. Broadly:
- Budget ($80–$150/mo). Usually rate-limiting plus a small scrubber. Suitable for game servers that attract occasional nuisance attacks, not for anything that has made an enemy.
- Mid ($150–$400/mo). Named scrubbing network, 100–500 Gbps declared, on-demand diversion, basic game protocol filters. This is where most dedicated game hosting sits.
- Enterprise ($400+/mo). Always-on, anycast, contractual unlimited mitigation, 24/7 NOC with a phone number, custom L7 filters, and traffic replay for forensics.
Protection tiers and what each one includes are compared in more detail at bestdedicatedwebhostingserver.com’s hosting comparisons; look at the gap between “declared capacity” and “written guarantee,” because that is where providers hide.
Configuration You Should Control Yourself
Provider mitigation is the outer wall. Inside it, there are controls that materially reduce your attack surface and are under your control:
- Do not expose the origin IP. A game server behind Cloudflare or a TCP proxy for web services should never reveal its real address in DNS, forum posts, or a misconfigured mail server. Use a separate IP for outbound mail.
- Kernel tuning for SYN floods.
net.ipv4.tcp_syncookies=1,net.core.somaxconnandnet.ipv4.tcp_max_syn_backlograised,net.netfilter.nf_conntrack_maxsized against your actual connection count. Defaults are sized for a web server, not a game backend. - Per-IP connection limits. An nftables rule limiting concurrent connections per source to 20–30 for game ports catches the low-and-slow rotation attacks that don’t trigger volumetric detection.
- UDP amplification hardening. Disable or restrict any UDP service that responds to reflection (game query ports that echo large responses are a common amplifier).
Questions to Ask Before You Sign
- Is mitigation on-demand or always-on, and what is the measured detection time?
- What happens at 120% of declared capacity — null route, overage bill, or absorbed?
- Where are the scrubbing centres relative to the server’s region, and is anycast used?
- Can you filter at Layer 7 for this specific game protocol, and how many times have you done it?
- Do you alert me when an attack is mitigated, via API or webhook as well as email?
- Is mitigation written into the SLA as unlimited, or does it have a numeric ceiling?
- What is the exit path if mitigation fails — can I get my IPs moved quickly?
If a provider can answer all seven in writing, the protection is real. If the answer to question two is vague, the protection is a marketing tier, and you should assume you are buying rate limiting and nothing more.




Leave a Reply
You must be logged in to post a comment.