DDoS scrubbing center: cleaning traffic away from your network
A DDoS scrubbing center is a site, run by a network operator or a mitigation provider, that receives another network's traffic during an attack, drops what looks like attack traffic and sends the rest on. The idea is old and simple: the flood is absorbed where there is more capacity than the victim could ever buy. The cost is a detour, a dependency on someone else's network, and a blind spot for anything that looks like an ordinary request.
Quick facts
- Core idea
- Divert traffic, filter it elsewhere, return the clean part (CISA)
- Diversion by BGP
- The protected address block is announced from the scrubbing network (RFC 4271)
- Diversion by DNS
- Names resolve to the provider's addresses; caches obey the TTL (RFC 1035)
- Two operating modes
- Always-on and on-demand
- Layer
- Network and transport; application requests only if the traffic can be read
- Signalling standard
- A victim can ask a mitigator for help with DOTS, described in RFC 8811
Scrubbing is one of several ways to mitigate an attack. The overall sequence is on DDoS mitigation; where scrubbing sits among other classes of solution is on cloud vs on-premise protection. This page covers the scrubbing step itself.
What a scrubbing center does
CISA describes the pattern in one sentence: a protection service detects abnormal traffic flows and redirects traffic away from your network, the attack traffic is filtered out, and clean traffic is passed on to you (CISA, understanding denial-of-service attacks; the page is archived, so treat its wording as general guidance). A scrubbing center is where the filtering happens.
- Divert. Packets addressed to the protected network are steered to the center instead of straight to the target.
- Scrub. Equipment there compares traffic against rules and baselines and drops packets that fit an attack: spoofed sources, malformed packets, floods of a protocol the target never uses.
- Return. What is left is forwarded to the real destination, often through a tunnel or a private interconnect.
The reason for doing this somewhere else is the one RFC 4732 gives for bandwidth-consumption attacks: they work by clogging the link to the victim. Once a link is full, nothing downstream of it, including a firewall in your own data centre, gets a chance to look at the packets. Filtering has to happen before the link, which means on someone else's capacity. The attack families this addresses are on volumetric attacks.
How traffic reaches it: BGP and DNS
Two mechanisms are common. They differ in what they can redirect.
Diversion by BGP announcement
BGP is how networks tell each other which address blocks they can reach; in the words of RFC 4271, the primary function of a BGP speaking system is to exchange network reachability information with other BGP systems. An UPDATE message advertises destinations as IP prefixes. To divert by BGP, the scrubbing network announces the customer's prefix. The rest of the internet then sends packets for those addresses to the scrubbing network, and the clean traffic is carried onward to the customer. A related use of the same protocol spreads load across sites: anycast.
The unit here is an address block, not a hostname. It covers every service in the block, including ones that are not web servers. It also needs a prefix the customer is allowed to announce, and prefixes that are too small are not accepted by other networks. Providers set their own minimum prefix size, so check the limit for any service you evaluate.
Cleaned traffic goes back to the customer's origin, commonly over tunnels or a direct interconnect. Setups differ in whether only inbound traffic passes through the provider or the replies do too. The return path is a design choice to settle with the provider.
Diversion by DNS
The other method changes where a name points. The customer's record is changed, or already points, to addresses owned by the scrubbing provider, so clients that look up the name connect to the provider. Only traffic sent to that name is diverted. Traffic aimed straight at an IP address, for instance one an attacker found earlier, is not. This is why the origin address has to be kept private and, ideally, closed to everything except the provider; the point is made again on servers and VPS.
DNS diversion is paced by caching. RFC 1035 defines the TTL as the time a record may be cached before the source should be consulted again, so a changed record reaches clients only as old copies expire. A low TTL set well before an attack shortens the wait; a long one set years ago does not.
A related tool: blackholing
Dropping all traffic to the attacked address is not scrubbing, because nothing is returned. It is covered on blackhole routing.
Always-on and on-demand
The same center can be used in two ways.
Always-on. Traffic passes through the center at all times. When an attack begins nothing has to change; the filters were already in the path. The costs are continuous: the fee, and the detour for every packet, attack or not.
On-demand. Traffic goes straight to the target until someone decides to divert it. The usual workflow is to advertise the prefix to the provider during an attack and withdraw it when the incident ends. Quiet periods cost less and add no distance. The weak point is the start. Detection has to notice the attack (DDoS attack detection), a person or a system has to trigger the change, and the change has to spread. The prefix documentation cited above puts global BGP convergence after an announcement from your own network at roughly five to ten minutes, and says the provider has no control over how fast other ISPs refresh routes. Those minutes are an attack aimed at an unprotected target.
Some designs reduce the gap by triggering diversion automatically. The DOTS architecture in RFC 8811 defines a signal channel through which a client asks a server for help in mitigating an attack, so detection on the customer side can start mitigation without a phone call. Whether a given provider supports it is a question for the provider.
Limits of scrubbing
- Latency. Diverted traffic travels to the center and then on to you, a longer path than the direct one. How much longer depends on where the center is relative to your users and your origin. This follows from the routing, and no figure applies to every service; ask a provider where its centers are and how the clean traffic returns.
- Capacity. A center is large, not unlimited. Its capacity is shared across customers, and the numbers a provider quotes are claims you cannot check from outside. Ask for them in a contract, not a brochure.
- Application layer. Scrubbing works on packets. A request that is syntactically correct and arrives at a normal rate looks the same as a customer's request. Inside HTTPS, the center sees nothing at all unless it terminates the connection with your certificate, a decision to make before an attack. The comparison table is on cloud vs on-premise, and the layer split is on layer 3, 4 and 7 attacks.
- Dependency. Routing or DNS now runs through a third party. If the arrangement is misconfigured, or the origin address is exposed, traffic can bypass the center entirely or stop arriving.
- False drops. Filters that are too tight remove real users. Watch customer-facing errors while scrubbing is active, not only attack volume.
Choosing a scrubbing service
Questions that decide most cases, none of which needs a vendor ranking: Does it divert by BGP, DNS, or both, and which one fits your address space? Always-on or on-demand, and how long from decision to effect? Where does clean traffic come back, and does it add an inspection point or only a tunnel? What does it do with HTTPS? Who triggers diversion, and at what hour? What happens to traffic that bypasses it?
Write the answers down with the contact details, test the diversion path while nothing is wrong, and keep a plan for undoing it. The steps after diversion are covered on DDoS mitigation; the first hour of a live attack on incidents.
Where a WAF helps and where it does not
Bridge WAF is not a scrubbing center, and this page does not claim that Bridge WAF or Bridge CDN offers one. A WAF works one layer up. It looks at HTTP requests that have already arrived and decides, per request, whether they match a rule. It cannot see packets that never became requests, and it cannot make a full network link bigger.
What it can do is deal with the part scrubbing struggles with: valid-looking requests aimed at expensive paths. Bridge WAF has three modes:
off
Nothing is inspected. Requests reach your origin as they otherwise would.
monitor
Matches are recorded and every request is still served.
block
Matching requests are refused and still recorded.
Start in monitor, read what would have been refused, and switch to block only when the matches are the traffic you meant to stop (WAF monitor mode). The activity log records what the filter flagged; it is not a view of attack volume and will not show a flood that was dropped before it reached the filter. For per-client caps on request rates, see rate limiting. For the application-level attack family, see application layer attacks.
Questions
What is a DDoS scrubbing center?
A DDoS scrubbing center is a facility with a large amount of network capacity and filtering equipment. Traffic for a protected network is diverted to it, attack traffic is dropped, and the remaining traffic is forwarded on to the real destination.
How does traffic get to a scrubbing center?
Mostly in one of two ways. The protected network's address block is announced over BGP from the scrubbing provider, so packets arrive there, or DNS records are changed so that clients resolve a hostname to the provider's addresses. BGP diversion covers whole address blocks; DNS diversion covers only names.
Is always-on or on-demand scrubbing better?
Always-on keeps traffic passing through the center all the time, so there is no switch-over when an attack begins. On-demand diverts traffic only after an attack is noticed, which costs less in quiet periods but leaves a gap while detection and route or DNS changes take effect.
Does a scrubbing center stop application-layer attacks?
Only partly. A center that sees packets can drop floods and malformed traffic. Requests that look valid, especially inside encrypted HTTPS, need a layer 7 filter that can read them, such as a WAF or rate limits at the application.
Does Bridge WAF run a scrubbing center?
No. Bridge WAF is the application-layer filter in front of your origin, with three modes: off, monitor and block. This page describes scrubbing as a class of service in general, not a feature of Bridge WAF or Bridge CDN.