Preparing Before the Flood: Capacity, Contacts, and Scrubbing
The difference between an organization that has a bad afternoon and one that has a bad month is decided before the flood arrives. During a denial-of-service attack, a small team can do surprisingly little on its own equipment. The response is mostly a matter of knowing what you have and knowing who to call. It is also a matter of having already made the decisions that are impossible to make well under pressure. None of that requires a security budget. Most of it requires an afternoon, a spreadsheet, and a few uncomfortable questions.
This guide is that afternoon. It walks through knowing your ceilings — the three numbers that tell you where each attack family would hurt first. It then explains the protection options in plain language, from the free ones to the ones that cost real money. It covers the contact sheet and the test call, tabletop drills for each family, and the monitoring alerts that turn a surprise into a notification. It also covers an honest cost-and-benefit discussion for organizations that cannot protect everything and should not try.
If the three families are new to you, DDoS in plain language explains them. The short version is that volumetric attacks fill your circuit, protocol attacks fill your connection tables, and application attacks bury your server. Preparation is about making sure each of those has a number, a plan, and a phone number attached.
In this guide
- Knowing your ceilings
- Where each family would hurt first
- Protection options in plain language
- The contact sheet and the test call
- Tabletop drills for the three families
- Monitoring readiness
- The honest cost and benefit
- The preparedness checklist
- Where to go next
Knowing your ceilings
Every resource an attack can exhaust has a limit, and the first act of preparation is writing those limits down. Three numbers cover most small networks.
Circuit capacity
Record the rated speed of each internet circuit, inbound, in bits per second. Separately, record the rate at which your edge router can actually forward packets. The second number is less obvious and worth asking your vendor about. A small router can be overwhelmed by a packet-rate flood well before its circuit is full. Also note what shares each circuit. If the public web server, the phones, the VPN, and office browsing all ride one link, a flood aimed at any of them takes all of them.
Firewall session limits
Record the maximum number of concurrent sessions the firewall can track and the rate of new sessions per second it can set up. Record the current timeout for half-open connections. The product of new-session rate and half-open timeout is roughly how many unfinished connections it takes to fill the table. That tells you how much of a protocol attack you can absorb. Read the maximum from the device rather than the brochure, since licensing and model variants change it.
Application capacity
The request rate at which each public application's response time starts to degrade, and the rate at which it fails. You can find this from history — the busiest hour you have ever recorded and how the server behaved. Or, better, use a modest load test against a staging copy, run by you, against your own system. In that test, the expensive endpoints (search, login, report, cart) are tested separately from the cheap ones. The number you want is not "how many page views can we serve" but "how many searches per second before the database is the bottleneck."
Two smaller items belong on the same page. An inventory of every public address and the service behind it, because you cannot decide what to sacrifice without knowing what is exposed. And the current time-to-live on each public DNS name, because a long TTL makes it slow to move a service behind protection when you need to.
Where each family would hurt first
With the numbers written down, the mapping is direct, and it is worth doing explicitly, because the answer is often surprising. A common small-organization result looks like this:
| Family | Your ceiling | What fails first | Who else is affected | Local levers |
|---|---|---|---|---|
| Volumetric | Circuit rated speed; router packet rate | The circuit, then everything behind it | Everyone sharing the link: office, phones, VPN | None that reduce the flood; only shedding and the provider |
| Protocol / state | Firewall max sessions; new sessions per second; half-open timeout | New connections through the firewall | Anyone starting a new session; established sessions may survive | Shorter half-open timeouts, per-source limits, firewall protections for half-open connections |
| Application | Requests per second on the expensive endpoints | The application, then the database behind it | Users of that application only, unless it shares a server | Rate limits, caching, disabling expensive features, a protective front |
The surprising row is usually the first one. Many small organizations discover that a flood aimed at a brochure website nobody would miss for a day would also take out the phones and the warehouse. That is because they all share one circuit. That discovery, on a calm afternoon, is the point of the exercise. It drives the most valuable protection decision in the next section.
Protection options in plain language
Protection is not one product. It is a short menu, and most small organizations end up with two or three items from it. Here is each one described by what it does and what it costs you, with no vendor names, because the categories outlive any particular offering.
Separating public services from internal circuits
The single most effective architectural move, and often the cheapest. Suppose your public website and customer-facing applications live on a different circuit. Or, more commonly, they live in a hosting environment or cloud service somewhere else entirely. In those cases, a flood aimed at them cannot fill the pipe your office, phones, and VPN depend on. The attack still hurts the public service, but it stops being an organization-wide outage. Conversely, if nothing public lives on your office circuit, that circuit is a far less attractive target. For many small organizations, moving the public web server off the office link is the whole DDoS strategy, and it is a good one.
Upstream provider protection
Your internet provider can filter, rate-limit, blackhole, or reroute traffic before it reaches your circuit. The guide on when and how to call the ISP covers each option and the call itself. As a preparation item, ask the provider what they offer, what it costs, how it is triggered, and how fast it happens. Write the answers on the contact sheet. Some providers include basic mitigation; some sell it; some can only blackhole. Knowing which is yours is free and takes one conversation.
Cloud scrubbing services
A scrubbing service is a facility with very large capacity that accepts all traffic for your addresses. It discards the attack and forwards the clean remainder to you. It can be always-on, where your traffic permanently flows through it. Or it can be on-demand, where you divert traffic to it when an attack starts. That usually happens by a DNS change or a route announcement. Always-on costs more and adds a little latency all the time. On-demand costs less but takes minutes to activate and depends on you noticing the attack. Either way, this is the option that keeps a service online through a large flood, and it is priced accordingly. It fits organizations with a public service whose downtime is expensive by the hour.
Fronting public services with a protective edge
A distributed front — the general category that includes content delivery networks and similar edge services — sits between the internet and your web server. Visitors talk to the front, which has enormous capacity spread across many locations. The front talks to your server, which is now called the origin and whose address should be hidden. Floods land on the front and are absorbed. Application-layer attacks can often be filtered there by rate and pattern. This is often the most cost-effective protection for a public website, with two cautions. It protects only what is behind it, so a mail server or VPN on the same circuit gains nothing. And if the origin's real address leaks — through an old DNS record, an email header, or a certificate — the attacker can go around the front. Checking that it has not is part of preparation.
Right-sizing tables and timeouts
The free option. Shorten the half-open connection timeout on the firewall to something reasonable. Enable whatever the device offers for holding unfinished handshakes cheaply. Set a per-source connection limit generous enough for real users. And raise the web server's connection backlog. These do not stop an attack; they raise the amount of attack you can absorb before the tables fill. Make the changes on a calm day, one at a time, and watch for a week to be sure nothing legitimate broke. The general guidance on session-table sizing lives in session-table monitoring.
Lowering DNS time-to-live in advance
If your plan includes moving a public service behind a protective front or scrubbing service during an attack, that move happens by changing a DNS record. The change propagates only as fast as the record's TTL allows. Lowering TTLs on public names to a few minutes, well before any attack, costs almost nothing and turns an hours-long move into a minutes-long one.
The contact sheet and the test call
During an attack, the network is down. The wiki that holds the phone numbers is behind the VPN. And the one person who knows the account manager's mobile is on holiday. The contact sheet exists to make that scene impossible. It is one page, kept in three places. Keep a printed copy near the firewall and a copy on a phone that does not depend on office Wi-Fi. Keep a copy inside the DDoS runbook.
| Contact | What you need on the sheet | Why |
|---|---|---|
| Internet provider | Emergency or operations number, abuse contact, account manager, every circuit ID and account number, what mitigation they offer and how it is triggered | The only party who can shrink a volumetric flood |
| Hosting provider | Emergency line, account number, what protection is included, whether they can move or shield your service | Public services hosted elsewhere have their own upstream |
| DNS provider | Login location, who holds the credentials, current TTLs, how to change records fast | Moving a service behind protection is a DNS change |
| Scrubbing or edge service | Activation procedure, support line, what is on-demand versus always-on | Only if you have one; if not, note that explicitly |
| Management | Who can approve spending, who can approve sacrificing a service, who speaks to customers | Decisions during an attack need named owners |
| Incident-response help | Retainer or insurer hotline, law enforcement contact for extortion | The hand-off point when it exceeds a small team |
Then make the test call. Dial the provider's emergency number on an ordinary afternoon. Say it is a contact verification. Confirm the number reaches people who know what a flood is and that your circuit ID means something to them. Do the same for the hosting provider. Note the date of the test on the sheet, and repeat it whenever a provider or an account manager changes. The general practice of holding providers to their commitments with evidence is covered in escalation with data. The test call is the DDoS-specific piece of it.
Tabletop drills for the three families
A tabletop drill is a conversation, not a test. Nobody touches equipment. A facilitator reads out a scenario in stages. The team says what they would look at, decide, and do. Someone writes down every point where the answer was "I don't know." Three drills, one per family, take about forty-five minutes each. The gaps they expose are the most useful output of this whole guide.
- Set the scene. "It is Tuesday at half past two. The monitoring system alerts that inbound on the main circuit has been above ninety percent for three minutes. What do you look at first?" Let the team answer before reading the next stage.
- Add the graphs. Describe what the six key graphs show for the family being drilled. For the volumetric drill, inbound is pinned and outbound is flat. For protocol, sessions are climbing with bandwidth calm. For application, response time and request rate are up with logins flat. Ask: what family is this, what is the target, and how do you know?
- Force the decisions. "The website is down and the phones are breaking up. Do you sacrifice the website to save the phones? Who approves that?" "The provider offers to blackhole the address in five minutes or set up a filter in an hour. Which do you ask for?" "A message arrives demanding payment. What happens to it?"
- Test the sheet. "Call the provider." Have someone actually pick up the printed contact sheet and read out what they would say, using the info sheet format. Time how long it takes to fill in the numbers.
- Record the gaps. Every hesitation is a finding: a missing number, an unknown ceiling, an unowned decision, a TTL nobody checked. Assign each one to a person with a date.
- Repeat the drill for the other two families in later sessions. Run all three again after any major change to the circuit, firewall, or public services.
Tip: the most valuable tabletop moment is the awkward one, when the facilitator asks "who approves sacrificing the website?" and two people look at each other. That silence, on a Tuesday afternoon with coffee, is a finding. The same silence at three in the morning with the director on the line is an outage extension. Drills exist to move the silence to the afternoon.
Monitoring readiness
The response playbook assumes that when the attack begins, someone finds out within minutes from a graph rather than within an hour from a customer. That requires a small number of alerts, tuned against your own baseline, that fire on the signatures of the three families. The recognition guide in this pillar explains the signatures. Here is what they look like as alert rules, in the plain-language form most monitoring systems can express:
# Volumetric: circuit approaching its ceiling WAN inbound utilization > 85% of rated speed sustained 3 min -> page WAN inbound packets/sec > 5x same-hour baseline sustained 3 min -> page WAN interface discards > 0 and rising sustained 3 min -> warn # Protocol / state: session table filling while bandwidth is normal Firewall sessions > 70% of maximum sustained 3 min -> page Firewall new sessions/s > 5x same-hour baseline sustained 3 min -> warn AND WAN inbound < 50% of rated speed (adds "bandwidth normal") # Application: server working harder for no more users Web median response time > 3x baseline sustained 5 min -> warn AND request rate > 2x baseline AND unique sources > 3x baseline AND logins per minute <= baseline (real users flat) -> page
The exact multipliers are starting points, not truths. Tune them until the alerts fire on drills and quiet days alike only when you would genuinely want to be interrupted. Three supporting items matter as much as the rules. Enable flow export from the edge device and keep it, so the top-sources view exists when you need it. Keep enough history to compare against the same hour last week. And route these particular alerts to a phone, not only to an email inbox that lives behind the circuit under attack.
The honest cost and benefit
A small organization cannot protect everything, and should not try. The honest framing is a short table you build yourself. For each public service, record what an hour of downtime costs, in money or in harm. Record what protecting it would cost and the cheapest option that meaningfully reduces the risk. The answers are usually lopsided in a useful way.
Protect what is public and revenue-bearing first. Take a customer portal that takes orders, a booking system, or a service customers pay to reach. These justify a protective front, a low TTL, and possibly on-demand scrubbing. A brochure website that nobody would miss for a day does not justify an always-on scrubbing subscription. It justifies being hosted somewhere other than the office circuit, behind a front if one is cheap, and otherwise being on the sacrifice list. Internal services should not be reachable from the internet at all, which is protection by absence. And the office circuit itself is best protected by having nothing on it worth attacking.
Then write down the decisions, including the ones to accept risk. "We will not pay for scrubbing. If the marketing site is flooded we will ask the provider to blackhole it and wait." That is a legitimate plan when it is written down and approved by the person who owns the marketing site. It is not legitimate as an unspoken assumption discovered during the incident. Preparedness is as much about agreed-upon sacrifices as it is about protection.
The preparedness checklist
| Area | Item | Done? | Owner and date |
|---|---|---|---|
| Ceilings | Circuit rated speed and router packet-rate limit recorded for each circuit | ||
| Ceilings | Firewall max sessions, new-session rate, and half-open timeout recorded | ||
| Ceilings | Application capacity on expensive endpoints measured or estimated from history | ||
| Ceilings | Inventory of public addresses and services; which circuit each shares | ||
| Ceilings | Family-by-family "what fails first" mapping written and reviewed with management | ||
| Protection | Public services separated from the office circuit, or a decision recorded not to | ||
| Protection | Provider asked the five questions; answers on the contact sheet | ||
| Protection | Protective front or scrubbing chosen for revenue-bearing services, or risk accepted in writing | ||
| Protection | Origin address confirmed hidden behind any front in use | ||
| Protection | Firewall timeouts and per-source limits right-sized; web server backlog raised | ||
| Protection | DNS TTLs on public names lowered to a few minutes | ||
| Contacts | One-page contact sheet complete; stored printed, on a phone, and in the runbook | ||
| Contacts | Test call made to provider and hosting; date recorded | ||
| Contacts | Decision owners named: spending, sacrifice, customer communication | ||
| Drills | Tabletop run for volumetric, protocol, and application; gaps assigned | ||
| Monitoring | Alerts for inbound saturation, session-table growth, and response-time-with-source-diversity in place and tuned | ||
| Monitoring | Flow export enabled; baseline history retained; alerts routed to a phone | ||
| Runbook | DDoS runbook written from the playbook skeleton; reviewed after each drill |
Where to go next
You now have the preparation that makes the rest of this pillar work. You have three ceilings written down and a mapping of where each family would hurt first. You have a menu of protection options with their honest costs and a contact sheet that has actually been dialed. You have drills that surfaced the awkward questions on a calm afternoon and alerts that notice the flood before customers do. And you have a set of written decisions about what to protect and what to sacrifice. The response playbook is what your team will follow when those alerts fire. And recognizing an attack in your graphs is the skill the alerts hand off to. Everything in this series lives under DDoS Detection & Response.
Put this into practice with SysPeep Monitor
Monitoring readiness is the part of preparedness you can finish this week. SysPeep Monitor discovers your edge devices and servers. It collects WAN utilization, packet rates, firewall session counts, and response times over SNMP and WMI from one Windows machine. It keeps the baseline history and sends threshold alerts by email and other notifications. So the rules in this guide become real alerts rather than good intentions. The free trial is a quick download.
Download Free TrialLearn About SysPeep