DDoS Response Playbook for Small Teams
It is the middle of a working afternoon. The public website has stopped answering. The phones are breaking up. And the inbound graph on the firewall is a flat line at the top of the chart. Two people are standing behind your chair. One of them is your manager. Nobody has done this before, and the one thing everyone agrees on is that you should do something, immediately.
This guide is what to do. It is written for a team of one to three people with no dedicated security staff, no scrubbing contract, and a firewall of the ordinary kind. It gives the first hour a shape — confirm, scope, protect, communicate, engage — so that the pressure to act turns into a sequence rather than a scramble. It also spends real time on what not to do. The most expensive mistakes in a denial-of-service incident are made by capable people acting quickly on the wrong instinct.
If you have not read them, DDoS in plain language explains the three attack families. The guide on recognizing an attack in your graphs covers how to tell one from a bad day. This playbook starts the moment you have a working hypothesis and needs to be usable at three in the morning by someone who is not you.
In this guide
- The shape of the first hour
- Confirm: attack, family, target
- Scope: what is down and what still works
- Protect what matters
- Communicate: honest and short
- Engage upstream
- Keep a timeline as you go
- What not to do
- The runbook skeleton and checklist
- When to hand off to outside help
- Where to go next
The shape of the first hour
Follow five verbs, in roughly this order. First, confirm what you are dealing with. Next, scope what is actually affected. Then protect the things that matter most. Next, communicate to the people waiting on you. And engage whoever upstream can do what you cannot. They are not strictly sequential. Communication and upstream engagement run in parallel with everything else, and you will loop back to confirm and scope as the picture changes. But the order matters for a reason: every step feeds the next. You cannot scope an attack you have not identified, cannot protect sensibly without a scope, and cannot communicate honestly or brief a provider without both.
If there are two of you, split immediately. Put one person on the graphs and the equipment, and one person on the timeline, the phone, and the status updates. If there is one of you, the timeline still gets kept — a text file with timestamps, one line per event. It is the only thing that will let you undo your own changes later.
Confirm: attack, family, target
The first ten minutes answer three questions, and you should not skip ahead until you can say all three out loud.
- Is it an attack? Rule out the look-alikes: a backup job pinning the circuit, one broken client in a retry loop, a partner's scanner, a marketing campaign that worked. The differentiating checks are in the recognition guide. The short version is that real traffic keeps relationships (inbound with outbound, requests with completed transactions) and attacks break them.
- Which family? Say it in resource terms. "The circuit is full" is volumetric. "The firewall session table is at its limit and bandwidth is normal" is protocol. "The web server is maxed and the network is calm" is application. The family decides who can help, so it is worth being precise.
- What is the target? Which of your public addresses, and which port or service, is receiving the traffic? This single fact drives scoping, protection, and the provider conversation. Read it from the top-destination view in your flow data or firewall logs.
Before you touch anything, capture the evidence: screenshots of the bandwidth, packet-rate, session, and application graphs with time axes showing, plus the top-sources and top-destination views. Once you start changing things, the picture changes with you, and you will want the original.
Scope: what is down and what still works
"Everything is down" is almost never true, and finding out what is not down is how you find out what to protect. Sort your services into three buckets — down, degraded, fine — and test them the way a user would, from outside your network if you can. A phone hotspot is the fastest outside vantage point a small team has.
Two scoping questions matter more than the rest. First: is the damage limited to the target, or is it collateral? If the attack is aimed at your web server but your office internet, VPN, and phones are also failing, they are sharing a saturated circuit. In that case, the real problem is the pipe, not the server. Second: is the attack on one address or your whole range? If only one public address is affected, sacrificing it is an option; if the whole range is under load, it is not. Write the answers down. They go straight into the status update and the provider ticket.
Protect what matters
Now decide what to defend. Not everything can be saved at once, and trying to save everything usually saves nothing. Pick the service that earns revenue, keeps people safe, or keeps the organization able to operate, and defend that first. Everything else is a candidate for shedding. With a priority chosen, the options below run from cheapest and most reversible to most disruptive.
Simple rate and connection limits at the edge
Most business firewalls can limit the number of connections a single source may hold. They can cap the rate of new connections to a given destination and shorten the timeout for half-open connections. Against protocol and application attacks, these are your best local levers. Apply them to the target service only. Set them generously (you want to stop a source holding a thousand half-open sessions, not a customer holding six). Note the exact time you enabled each one. Against a volumetric flood they will not help — the circuit was full before your firewall saw the packets. You should not spend more than a minute confirming that before moving to the upstream step.
Temporary filtering, where sensible and reversible
Suppose the source view shows a clear pattern: one service port, one protocol you do not use, one region your organization has no customers in. In that case, a temporary access-control rule that drops it at the edge can buy relief for a protocol or application attack. Be honest about the costs. Region-based blocking cuts off real customers and travelling staff, source addresses may be forged, and any list you build by hand will be stale in minutes. Treat these rules as tourniquets: apply them deliberately, label them with the incident and time, and set a reminder to remove them. Never block your own monitoring, your provider's addresses, or the ranges your remote staff use.
Moving the target behind a protected front
Suppose you already have an arrangement that can front the public service — a scrubbing service, a protective edge, or a hosting provider with absorption capacity. Now is when you use it, usually by pointing the service's public name at the protected address. Two cautions. DNS changes take effect only as fast as the record's time-to-live allows. So a record with a long TTL will keep sending users to the old address for hours. This is why preparation includes lowering TTLs on public names in advance. And the old address is still receiving the flood, so the move helps only if the attacker is following the name rather than the address. This option is close to useless if the arrangement does not already exist; it is not something to shop for during the incident.
Deliberately shedding a service
Sometimes the most protective move is to take the target offline yourself. That might mean a static "we are experiencing issues" page or a disabled search feature. Or it might mean an application put into a reduced mode that refuses the expensive operations. Shedding a non-essential service to keep the essential one alive is a legitimate decision, not a defeat, and it is far better made on purpose than by collapse.
Communicate: honest and short
While the technical work continues, someone has to tell management and users what is happening. The guide on incident communication covers the craft in depth; the DDoS-specific points are these. Say what is affected, what still works, what you are doing, and when the next update will arrive. Do not say "attack" to the public until you are sure; "abnormal inbound traffic" is accurate and does not commit you. Do not promise a resolution time. Do send the next update when you said you would, even if it says "no change." Silence is what turns a bad hour into a bad reputation. A first notice fits in three sentences:
14:22 - Our public website and customer portal are unavailable due to abnormal inbound traffic affecting our internet connection. Internal systems and email are working; phones may be degraded. We are working with our provider and will update by 14:50.
Management gets one more sentence than users. Include which family you believe it is, what you have asked the provider for, and what decision you may need from them. That decision might mean approving a temporary block of a region, or accepting that one address will be sacrificed to protect the rest.
Engage upstream
The rule is simple: if the circuit is saturated from outside, only someone upstream of it can help. No local change reduces traffic that has already been dropped before reaching you. The moment you confirm a volumetric attack, the provider call becomes the highest-value thing anyone on the team can do. That call should not wait for the protection steps above. For protocol and application attacks, engage upstream in parallel with your local levers if the attack persists or grows, since providers can also filter patterns you cannot.
Have the target addresses, the start time, and current inbound bit and packet rates ready when you dial. Also have the source pattern you see, the family you suspect, and your circuit identifier. The dedicated guide on when and how to call the ISP walks through the call, what they can do, and the tradeoffs of each option. Keep it beside the phone.
Keep a timeline as you go
A timeline is a plain text file, chat channel, or notebook page with one line per event. Each line has the time, what you observed or did, and any numbers. Keeping the timeline is the least glamorous task in the incident and the one you will be most grateful for afterwards. The timeline lets you undo your own changes in reverse order when the attack stops. It gives the provider a coherent history instead of a panicked memory. It becomes the after-action review, and the evidence for insurance or law enforcement if extortion is involved. It also becomes the answer to "what happened?" when the director asks next week. Keep it in the moment, not from memory afterwards; times remembered under stress are wrong by more than you would expect.
Tip: change one thing at a time, and write it down before you do it. Under pressure, the temptation is to enable four firewall features, add two block rules, and restart a service in the same two minutes. When the graphs then move, you will have no idea which change caused it — or whether the attacker simply changed tactics. One change, one timestamp, one look at the graphs, then the next.
What not to do
- Do not reboot the firewall reflexively. A reboot clears the session table, which feels like relief for about ninety seconds, until the attack refills it. In exchange you lose the current state, every established user session, the in-memory logs, and several minutes of visibility while the device comes back. Reboot only if the device has genuinely hung.
- Do not make many changes at once. See the tip above. Your future self needs to know which lever worked.
- Do not engage the attacker. If an extortion message arrives, do not reply, do not negotiate, and do not pay. Preserve the message with its headers, note the time, and hand it to management and, where appropriate, law enforcement. Replying confirms a live target.
- Do not promise a time. You do not control when a flood ends. "We are working on it and will update at 14:50" is a promise you can keep; "it will be back in an hour" is not.
- Do not disable logging to save CPU. The logs are your evidence and your provider's map. If a device is struggling, reduce what it logs, not whether it logs.
- Do not block your own lifelines. Every emergency rule should be checked against your monitoring addresses, your provider's ranges, and your remote staff before it goes live.
- Do not assume it is only a flood. A denial-of-service attack is occasionally cover for something else. While the pipe is full, keep an eye on authentication logs and anything that should never change during an outage.
The runbook skeleton and checklist
Everything above should live in a document that a colleague can follow without you. If your team has never written a runbook, your first five runbooks explains the format and how to write one quickly. This is the skeleton for the DDoS one.
DDoS RESPONSE RUNBOOK 0. Contacts: provider emergency line + circuit ID, hosting, DNS, management, IR help. (Verified on: ______) 1. Confirm: family? target address/port? evidence captured? -> one sentence in resource terms 2. Scope: down / degraded / fine. Collateral? One address or range? 3. Protect: priority service = ______ a. connection + rate limits on target (time enabled: ____) b. temporary ACL (pattern, time, removal reminder) c. move behind protected front (DNS TTL: ____) d. shed non-essential service 4. Communicate: first notice within 15 min; next-update time stated 5. Engage upstream: call placed at ____; ticket #____; ask = ______ 6. Timeline: one line per event, kept live 7. Stand-down: remove temp rules in reverse order; after-action within 3 days
The checklist below is the same material as a table you can print and tick. The "owner" column is there for a reason. In a two-person incident, saying who does what out loud prevents both of you doing the same thing and neither of you calling the provider.
| Phase | Action | Owner | Evidence or output |
|---|---|---|---|
| Confirm | Rule out look-alikes; name the family; identify target address and port | Tech lead | One-sentence hypothesis; graph screenshots |
| Confirm | Capture top sources, top destination, rates, start time | Tech lead | Saved views; timeline entry |
| Scope | Test services from outside; sort into down / degraded / fine | Second person or tech lead | Three-bucket list |
| Scope | Determine collateral vs. target; one address vs. range | Tech lead | Timeline entry |
| Protect | Choose the priority service; apply limits to the target only | Tech lead | Change log with times |
| Protect | Apply reversible filtering or shed a service if warranted | Tech lead, management approval | Labelled rules; removal reminder |
| Communicate | First notice within fifteen minutes; state next-update time | Second person | Sent message; timeline entry |
| Communicate | Brief management on family, ask, and pending decisions | Second person | Decision recorded |
| Engage | Call provider with the info sheet; get a ticket number and a named contact | Second person | Ticket, contact, agreed action |
| Throughout | Keep the timeline; one change at a time | Everyone | Timeline file |
| Stand-down | Remove temporary rules in reverse order; confirm graphs normal; final notice | Tech lead | Clean configuration; closing message |
When to hand off to outside help
A small team's playbook has to include the point at which it stops being a small-team problem. None of these criteria mean you failed; they mean the incident has exceeded what one or two people with a firewall can be expected to handle.
- The circuit is saturated, the provider has been engaged, and after their first action the flood is still landing. Ask the provider for their security or abuse team specifically, and ask about the rerouting or scrubbing options they offer.
- The attack changes shape when you respond — a volumetric flood becomes an application flood becomes a state attack. Adaptive attackers are a signal that you are outmatched on time and attention, not skill.
- An extortion demand has arrived. Management, legal counsel, your insurer's incident line if you have one, and law enforcement all belong in the conversation now.
- The attack has persisted beyond the point where the affected service's absence causes real harm, and you have exhausted your local levers. Hosting the public service somewhere with absorption capacity may be a same-day decision.
- You suspect the flood is cover for an intrusion. That is an incident-response engagement, and it should not wait for the flood to end.
- You are alone, it has been several hours, and you are making mistakes. Fatigue is a hand-off criterion too.
When you hand off, the timeline is the handover document. A page of timestamped observations and changes will let a provider's engineer or an outside responder pick up in minutes rather than re-deriving everything you already know.
Where to go next
You now have a first hour with a shape. Confirm the family and target, and scope the damage. Protect the service that matters most with reversible levers. Communicate honestly on a schedule, and engage upstream the moment the circuit is the problem. You have a list of the instincts to resist, a runbook skeleton to fill in, and the criteria for calling in help. Two guides in this pillar sit on either side of this one. When and how to call the ISP goes deep on the upstream step. The guide on preparing before the flood covers the contacts, capacity knowledge, and drills that make this playbook run smoothly instead of for the first time. Both live under DDoS Detection & Response.
Put this into practice with SysPeep Monitor
A playbook only works if the confirm step takes minutes, and that depends on graphs that already exist when the phone rings. SysPeep Monitor collects WAN utilization, packet rates, firewall session counts, and server response times from a single Windows machine. It keeps the history that shows what normal looked like, and alerts on thresholds. That way, the first ten minutes of this playbook are reading, not guessing. Try the free trial before you need it.
Download Free TrialLearn About SysPeep