Under attack?
Mitigation

Filtering where the packet lands, not where it hurts

Our filter is an eBPF program attached to the receive path of every network card in every cluster. A packet is classified and dropped before the kernel allocates a buffer for it. That is why a drop costs nanoseconds - and why capacity is measured in tens of millions of packets per second per server, and a billion per scrubbing location.

1000Mpackets per second filtered per scrubbing location - ten servers at 100M each, in a redundant cluster
4000Gof filtering capacity per scrubbing location - ten servers at 400G each, PCIe Gen6 on ConnectX-8
<1sfrom first attack packet to enforced rule
10European locations - six scrubbing clusters, four edge nodes
How it decides

Filtering where the packet lands, not where it hurts

Most providers inspect traffic after the operating system has already paid for it. We attach to the driver's receive path, so a dropped packet costs a few nanoseconds and never touches the stack.

L3 - network

Volumetric absorption

Amplification and reflection floods dropped at the nearest scrubbing cluster. Capacity is per cluster, not pooled across a region.

L4 - transport

Stateful TCP defence

SYN cookies, connection tracking and rate limits keyed on the TCP option fingerprint, so a flood is separated from a busy customer.

L7 - payload

Protocol validation

Payload judged on its bytes, never on its port number. A DNS reply is recognised as DNS wherever it arrives.

Coverage

What we stop

Every vector below has its own counter, its own threshold and its own drop reason. When something is mitigated you can be told which rule fired and why - not just that “an attack was blocked”.

Your traffic is never blackholed. A prefix under attack keeps being announced and keeps being reachable. Dropping a customer off the internet to save the network is not mitigation, it is the attacker’s objective reached by other means - and it is the one thing we have never done.

TCP

Transport and handshake

  • TCP SYN flood
  • TCP SYN+ACK flood
  • TCP ACK + PSH flood
  • TCP FIN flood
  • TCP RESET flood
  • Fragmented ACK attack
  • TCP fragments
  • Invalid TCP MSS
  • Invalid TCP WIN
  • Invalid TCP SEQ
  • Stateful check failures
  • Connection flood
  • Low-rate attacks
  • Booter TCP floods
  • LOIC and similar tooling
UDP

Floods and reflection

  • UDP flood
  • Carpet bombing across a prefix
  • UDP null packets
  • Invalid checksum
  • Invalid packet size
  • Invalid TTL
  • hping3 and booter scripts
  • Reflection & amplification
  • DNS · NTP · SSDP · CLDAP
  • memcached · chargen · SNMP
  • Portmap / rpcbind · MS-SQL
  • TFTP · RIP · BitTorrent · WSD
  • CoAP · NetBIOS
Other

Network layer and protocol abuse

  • IP spoofed packets
  • ICMP flood
  • ICMP port unreachable
  • Smurf attack
  • Ping of death
  • Teardrop attack
  • IP fragment floods
  • SIP attacks
  • IPsec attacks
  • Game protocols
  • Source engine query floods
  • RakNet / Minecraft floods
  • FiveM and voice-service floods
Second layer · separate service

Layer 7: a WAF and reverse proxy in front of the application

not part of the network filter

Everything above is decided in the network, on the packet, at line rate. An attack that is indistinguishable from a real user at the packet level cannot be. A valid HTTPS request is a valid HTTPS request whatever its intent. It has the right handshake, the right headers and a real TCP connection behind it. No amount of filtering in a NIC driver will tell it from the request beside it.

Judging that needs the decrypted request, which means terminating the connection. So the application layer is a second engine with a different architecture: a reverse proxy that accepts the session, inspects what was actually asked for, and forwards only what survives.

It sits behind the network filter, not instead of it. That ordering is the part most WAF products cannot offer. A proxy exposed directly to the internet has to survive the volumetric attack itself - and a 300 Gbps flood takes it off the air long before it evaluates a single request. Here the packet filter absorbs the flood first and the proxy only ever sees traffic that already passed, so the two layers never compete for the same machine’s resources.

TLS

Certificates issued and renewed for you

Certificates are generated automatically the moment a hostname is pointed at us, and renewed without anybody being asked. Issuing a certificate is easy and everyone gets it right; renewing one at three in the morning eighty-nine days later is what actually takes sites down, which is why that half is automated and not documented.

Challenge

A captcha only when something looks wrong

Suspicious sessions are challenged; everyone else is not. An always-on captcha is a tax on every real visitor to catch a few bad ones, and it trains people to leave. The challenge escalates with the evidence - most legitimate users never see one, and the ones who do see it once.

Client classification

Proxies, Tor exit nodes, bots and scripts

Anonymising infrastructure and automated clients are identified and handled by policy: open proxies, VPN egress, Tor exit nodes, headless browsers and scripted clients that imitate one. Policy, not verdict - for a shop, Tor traffic may be worth blocking outright; for a news site it may be the readers who need it most. You choose per hostname, and we do not choose for you.

Request filtering

What the application would rather not receive

Request-rate limits per client rather than per address, so one NAT does not stand in for a thousand users. Plus the ordinary WAF work: injection and traversal attempts, malformed and oversized requests, and login endpoints protected against credential stuffing by rate rather than by blocking a country.

We say all of this plainly because the alternative is selling one product and delivering another: several providers put “HTTP flood” on the same list as “UDP flood” without mentioning that the two are stopped by entirely different machinery - or by none.

  • HTTP flood
  • HTTPS flood
  • Slowloris and low-and-slow
  • Apache Killer
  • Brute force / credential stuffing
  • Buffer overflow attempts
  • Scripted and bot traffic
  • Proxy · VPN · Tor origin
  • Request-rate abuse
Getting connected

Two ways to put the filter in front of you

Protection is delivered over ordinary networking - a tunnel from wherever you already are, or a cable in a facility we share. No agents on your servers, no DNS tricks for the network layer, nothing to install.

From anywhere

GRE, IPIP or VXLAN tunnels

You stay at your current provider and nothing physical changes. We announce your prefixes (or you keep announcing them and steer traffic to us), inbound traffic crosses the filter, and what survives is delivered to your routers through the tunnel.

  • GRE and IPIP for routed hand-off; VXLAN where you want Layer 2 semantics preserved
  • Redundant tunnel endpoints in separate locations, with failover between them
  • MTU stated and tested before turn-up, not discovered in production
  • Return path stays yours: outbound traffic keeps flowing directly, asymmetry is expected
  • Turn-up is a configuration session, not a project - same-day is normal
In our locations

Cross-connect at any of our ten sites

If your equipment is in - or near - a facility we are present in, a cross connect replaces the tunnel: no encapsulation, no MTU overhead, and the filter sits directly in the physical path. This is also how protected IP transit is delivered.

  • Available at all ten European points of presence, 100G and 400G handoff at every one
  • 1G to 400G ports; optics per the path - 10GBASE-LR through 400G LR4
  • Cross connect ordered and arranged by us with the facility
  • Dual ports at one site or diverse ports across two, over EVPN-MPLS behind them
  • Combine with transit, a wavelength or Layer 2 transport on the same port

Both arrive at the same filter with the same thresholds - the delivery method changes where your clean traffic comes out, never how it is cleaned. Not sure which fits your topology? Ask an engineer - it is usually a five-minute answer, and a longer one is a service we also sell.

In combination

Protection rarely travels alone

Most protected customers end up holding two or three of these - not because we bundle them, but because each one removes a hand-off that used to be a weak point.

IP transit

The pipe and the filter, one supplier

Buy the connectivity from the same network that cleans it and the finger-pointing disappears: no diversion event, no third-party GRE, one NOC that owns the whole path. Protected IP transit →

LIR services

Protect addresses that are actually yours

Protection follows the prefix. With your own ASN and PI space the protected resources are registered to you - portable, permanent, and never hostage to any provider, including us. ASN & address space →

Consulting

Or build the architecture with us

Where filtering belongs in your topology, what it must see to decide anything, and which attacks a design provably cannot stop - answered before you spend, not after. How an engagement runs →

Frequently asked

How fast does mitigation start?

It does not start - it is already running. Filtering is permanently in the path, so there is no activation, no diversion and no BGP convergence to wait out. A new vector goes from first packet to enforced rule in under a second.

Will you ever null-route my prefix?

As our own mitigation step - never. Blackholing completes the attack, so our escalation path ends in a tighter filter, not in your prefix disappearing. RTBH exists as a tool you can trigger for your own reasons; we do not use it on you.

Do I need a GRE tunnel?

Only if your servers stay at another provider. Tunnels (GRE, IPIP, VXLAN) are one delivery method; a cross connect at any of our sites or protected transit are the others. Same filter either way - only where the clean traffic exits changes.

Does the filtering add latency?

Nanoseconds. The decision happens in the NIC driver before the kernel allocates memory for the packet, and there is no scrubbing-centre detour because there is no detour at all - the filter sits on the path your traffic already takes.

What do I get after an attack?

An incident record: which rule fired, what threshold it crossed, what the traffic was made of, and a packet capture you can open in Wireshark. The exact format is on the evidence page - we would rather you checked than trusted.

My traffic is unusual - game servers, VoIP, custom UDP. Will it survive?

That traffic is why the filter judges packets by their bytes, never by their port. Source engine, RakNet, SIP and RTP are recognised as themselves, and thresholds are set against your measured baseline rather than an industry default.