skip to content

Timing & Performance

Timing templates, rate limits and retries set how fast a scan runs and how much load it adds. Interviewers expect -T5 can give wrong results, and how a defender spots a slow or fragmented scan.

on this pageshow

explore

questions

6

In Nmap 7.991, what do the six -T timing templates from paranoid (-T0) to insane (-T5) change, and which one runs by default?

level: juniorimportance: must knowfreq 45%

answer

  1. six named levels, zero to five
  2. the default sits in the middle
  3. lowest three run one probe at a time
  4. bounds on delay, timeouts and retries
  5. higher numbers assume a reliable network

basics

~20 s

Nmap's -T option takes 0 to 5 or a name: paranoid, sneaky, polite, normal, aggressive, insane. -T3 normal is the default. -T0 to -T2 serialise probes with a delay between them; -T4 and -T5 shorten timeouts and retries for speed.

solid answer

~30 s

`-T` picks a bundle of timing settings by number or name: `-T0` paranoid, `-T1` sneaky, `-T2` polite, `-T3` normal, `-T4` aggressive, `-T5` insane. `-T3` is what runs when you give no `-T`, so passing it changes nothing. The three slow templates scan one probe at a time with a pause between probes (five minutes, fifteen seconds, 0.4 seconds). `-T4` and `-T5` keep Nmap's parallelism but shorten round-trip timeouts, cut retransmissions (6 and 2) and cap the dynamic scan delay (10 ms and 5 ms). An explicit option such as `--max-retries` overrides the template's value for that one parameter.

go deeper

for a junior

Recall the six names in order, that -T3 is the default, and that lower numbers are slower while higher numbers trade accuracy for speed.

for a middle

Explain what each end actually changes: serialised probes with delays at -T0 to -T2, shorter RTT timeouts and fewer retries at -T4 and -T5, and that explicit options win over the template.

for a senior

Choose the template from the target network's reliability and the window, and say why -T2 is not a free safety margin when version detection is the bigger load.

for a principal

Decide which template an assessment standard names as the default, and how a plan records pace so the receiving team can match it to the log.

## What a timing template is Nmap adapts its speed to the network as it scans: it measures round-trip times, widens or narrows how many probes are outstanding, and retransmits when a reply does not arrive. The fine-grained options (`--max-retries`, `--max-rtt-timeout`, `--scan-delay` and others) bound that behaviour. A **timing template**, set with `-T`, is a named preset of those bounds, so an operator does not have to pick each value. You can write it as a number (`-T4`) or a name (`-T aggressive`). ## The six templates in Nmap 7.991 | Template | Name | What it sets (reference guide and source) | |---|---|---| | `-T0` | paranoid | One probe at a time, five minutes between probes | | `-T1` | sneaky | One probe at a time, 15 seconds between probes | | `-T2` | polite | One probe at a time, 0.4 seconds between probes | | `-T3` | normal | The default: adaptive and parallel, no template overrides | | `-T4` | aggressive | RTT timeouts 100 ms min, 500 ms initial, 1250 ms max; max retries 6; max TCP and SCTP scan delay 10 ms | | `-T5` | insane | RTT timeouts 50 ms min, 250 ms initial, 300 ms max; max retries 2; host timeout 15 minutes; script timeout 10 minutes; max TCP and SCTP scan delay 5 ms | Without a template, the retransmission cap is ten (up to eleven probes to one port) and the dynamic TCP scan delay may rise as far as one second. ## How to read the table - **The first three are about spacing.** Serialising probes and waiting between them uses little bandwidth and gives a fragile host room, but the cost is time. At a five-minute gap, one host's default list of 1,000 ports takes at least about three and a half days; at 15 seconds, about four hours; at 0.4 seconds, a little under seven minutes before any other delay. - **The last two are about patience.** Short timeouts and few retransmissions make Nmap give up on silent ports sooner. That saves time, and it is also where accuracy is lost, because a slow reply is indistinguishable from no reply once the timer has fired. - **`-T3` is a baseline, not a mode.** The reference guide notes that it does nothing when you pass it, and that machine crashes and bandwidth problems are rare at the default. ## Choosing one in an authorised audit 1. Start at `-T3`, or `-T4` on a modern, reliable network. The reference guide recommends `-T4` on a decent broadband or Ethernet link and keeping it when adding fine-grained options, because it brings small optimisations that no single option reproduces. 2. Treat `-T2` with care. People choose it as a courtesy, but the reference guide says it can take ten times longer than a default scan, and that omitting version detection reduces load far more effectively than playing with timing values. 3. Leave `-T0` and `-T1` to cases where a very long, very sparse scan is the point. For a long scan the reference guide suggests setting the exact values instead of relying on the canned ones. 4. Say the template in the assessment's written plan, so the person reading the target's logs can match the observed pace to what was agreed. ## Confirming what actually ran A result is only interpretable with its settings. Two places hold them: - In **XML output**, the `nmaprun` element's `args` attribute records the command line, including any `-T` and fine-grained options. - With **debugging (`-d`)**, Nmap prints a *Timing report* block before scanning that lists the effective RTT timeouts, maximum scan delays, parallelism, retries, host timeout and the minimum and maximum rates. It is the quickest way to see what a template plus your overrides combined into. This matters when two runs disagree, or when someone on the receiving side asks which settings produced a given log: `-T4` with explicit overrides is not the same scan as plain `-T4`. ## Common mistakes - Reading the number as a speed multiplier. The templates change timeouts, retries and delays, not a single speed knob. - Using `-T0` or `-T1` on a long list of hosts and finding the scan cannot finish in the window. - Adding `-T5` to a scan across a wide-area link and trusting the quicker answer as complete. ## Precedence Templates and fine-grained options combine. The options you give explicitly take precedence over the template's default for that parameter, and the template's other settings stay. So `-T4 --max-retries 3` retries up to three times, but still keeps `-T4`'s timeouts and its 10 ms cap on the dynamic TCP scan delay. ## From the other side A defender reading logs sees the template as a pace: a handful of probes a second to dozens of ports at `-T3` or `-T4`, a steady trickle at the slow templates, a short intense burst at `-T5`. The pace is a property of the tool's configuration, not of intent, which is why pace alone never settles whether a scan was authorised.

  • A colleague runs -T2 across a production /24 'to be polite'. What does that cost, and what actually reduces impact more?
    The reference guide warns that polite mode can take ten times longer than a default scan, and that crashes and bandwidth problems are already rare at `-T3`. Omitting version detection reduces load far more effectively than playing with timing values. So the scan runs longer, over a bigger window, without a clear gain.
  • You run -T4 --max-retries 3. Which retry cap applies, and what else from -T4 survives?
    Three. Options you give explicitly take precedence over the template's value for that parameter. The rest of `-T4` stays: its 100/500/1250 ms RTT timeouts and its 10 ms ceiling on the dynamic TCP scan delay. The template is a preset of defaults, not a mode that rejects overrides.
  • Why is -T3 'normal' described as the default if the number sits in the middle of the scale?
    Because it is the behaviour with no template overrides applied: parallel probing with Nmap's adaptive timing and the built-in limits (ten retransmissions, up to a one-second dynamic TCP scan delay). The other five templates each replace some of those limits, so `-T3` is simply the absence of an override.

  • -T0 is the fastest because zero means no delay
  • -T3 is a slow cautious mode you must request
  • Higher templates only send packets faster; nothing else changes
  • -T2 polite finishes in about the same time as the default
  • A -T template overrides an explicit --max-retries

Assumes Nmap 7.991; the template values are read from the 7.991 reference guide and source.

open as a page

Why do filtered ports dominate an Nmap scan's duration, and what do --max-retries and --host-timeout give up to shorten it?

level: middleimportance: should knowfreq 24%

basics

~20 s

A port that never answers makes Nmap wait out a timeout and retransmit, up to ten times by default. --max-retries cuts the retries and risks missed ports; --host-timeout abandons a slow host entirely, with no port table, OS or version results for it.

open as a page

Why can -T5 or a high --min-rate cost accuracy in Nmap 7.991, and what do you do when two scan runs disagree?

level: middleimportance: should knowfreq 33%

basics

~20 s

-T5 shortens round-trip timeouts to 300 ms and allows only two retransmissions, so slow or lost replies are read as silence. A high --min-rate can do the same through congestion. Treat disagreeing ports as unproven and rescan them at default timing.

open as a page

An assessor with a signed scope must run an Nmap scan of a fragile production /24 without disturbing it. Which pacing options and scan choices do you set, and how do you confirm the limit held?

level: seniorimportance: should knowfreq 28%

basics

~20 s

Run the lightest scan that answers the question: no version detection or scripts, a canary first. Cap the global rate with --max-rate, space probes to fragile hosts, bound the run with --max-retries and --host-timeout, then check the observed rate.

open as a page

What do Nmap's -f and --mtu options send, and what does a defender's firewall or sensor see and do about those tiny IP fragments?

level: seniorimportance: nice to knowfreq 8%

basics

~20 s

-f splits probes into IP fragments of eight bytes or less after the IP header; --mtu sets a multiple of eight. A defender sees bursts of tiny fragments from one source; a firewall or sensor that reassembles them inspects whole packets.

open as a page

How does a defender recognise a slow Nmap scan that touches one port every five minutes for days, and what can an empty log not prove?

level: seniorimportance: nice to knowfreq 14%

basics

~20 s

A slow scan shows as one source touching many distinct ports or hosts at a steady pace, so count them per source over hours or days, not seconds. An empty log proves only what the control logged, sampled and kept.

open as a page