← all systems · see it in The System
Botnet
A botnet is a group of Internet-connected devices, each of which runs one or more bots, controlled by its owner through command-and-control software (src:wp:botnet). Walked as botnet researchers describe it: the population of compromised computers, phones and IoT devices; how bots are recruited (scanning and brute force, drive-by downloads, trojans, spam) at the level the measurement literature describes; the command-and-control rendezvous (IRC, web domains, domain generation algorithms, fast-flux, peer-to-peer); the herder's control; the uses and the market (DDoS, spam, credential theft, fraud, proxies, rental); the operators; the defenders who measure, sinkhole, detect and take botnets down; and their history from 2001 to the 2024-2025 takedowns.
Properties published scienceour proposal
Sourced fields, and our own readings.
| born at | 2001 (first acknowledged); Mirai 2016-08-01 |
| born precision | year |
| boundary | Inside: the compromised devices while they run the bot and answer the controller, the bot programs, the command-and-control rendezvous (servers, channels, domains, peer lists, proxy and loader machines) and the herder's control. Outside: the devices' owners and their other uses, the malware family as a kind (its own system), the operators as people (threat actor), the customers who rent it, the targets, the Internet and DNS it runs over, and the defenders, researchers and police who act on it. A bot cut off from C&C stays infected but sits at the edge: dormant, not commanded. |
| nested level | Built from compromised computers and IoT devices running bot malware, joined by a C&C channel; run by threat actors; carried by the Internet; a criminal infrastructure inside human society's cybercrime economy. |
| scale | technological |
Measures (29) published science
| characteristic clock: bot check-in interval (Torpig) | 20 minutes |
| Mirai doubling time (first infections) | 76 minutes |
| Mirai infections in its first 20 hours | nearly 65,000 devices |
| Mirai steady-state population | 200,000-300,000 (later 100,000; brief peak 600,000) devices |
| Torpig takeover length | 10 days |
| Torpig bots vs unique IPs (10 days) | 182,800 bots; 1,247,642 IPs count |
| Torpig DHCP churn factor (IPs per bot), all countries | 6.83 (1.75 NL to 13.35 DE) ratio |
| Torpig new IPs vs new bots per hour | 4,690 vs 705 per hour |
| share of Torpig bots behind NAT/VPN/proxy/firewall | 78.9 % |
| Torpig accounts stolen in ten days | 8,310 accounts at 410 institutions |
| Torpig data collected in ten days | almost 70 GB |
| Krebs on Security attack volume (Sept 2016) | over 600 Gbps |
| Mirai DDoS attack commands in five months | 15,194 commands |
| Mirai credential dictionary | 62 pairs (BASHLITE: 6 usernames, 14 passwords) |
| Conficker C DGA output | 50,000 domains per day over 113 TLDs; 30 organisations to contain domains/day |
| DGA-based malware families analysed | 43 (253 seeds) families |
| domain registration price that sets the cost of sinkholing | about 5-10 per year per .com/.net USD |
| crawler underestimate of P2P botnet size | up to two orders of magnitude; 60-87% of nodes unverifiable behind NAT factor |
| lifespan of active P2P botnets (2013) | as long as five years |
| largest P2P botnet families | over a million bots |
| Storm peers publishing Storm content per day | 5,000-6,000 peers/day |
| spam conversion (Storm pharmacy campaign) | 28 sales from almost 350 million mails in 26 days (well under 0.00001%) sales/mails |
| revenue of the measured campaign | 2,731.88 (over 100 a day) USD |
| share of world spam from zombies (2005 estimate) | 50-80 % |
| first acknowledged botnet's share of spam (2001) | nearly 25 % |
| one request per bot every ~10 minutes yields | more than 5 million attempts per day attempts/day |
| Operation Endgame, May 2024 | 4 arrests; 100+ servers; 2,000 domains count |
| DDoS scale by 2016 | over 1 Tbit/s |
| Sandia emulation | 1,000,000 Linux kernels on a 4,480-node cluster VMs |
Open contentions (3) open contention
Where sources or readings disagree. Both stay on the map.
- C-1595 How do you count the machines in a botnet?
booked: The map keeps each count with the method that produced it, since counting addresses, bot IDs or crawled peers can differ by ten to a hundred times.
the other reading: unique IPs (overcount ~an order of magnitude: Torpig 1,247,642 IPs vs 182,800 bots); bot IDs in C&C traffic; crawler enumeration of P2P (undercounts up to two orders); sensor injection - C-1596 Does taking down a botnet end it?
booked: The map books takedowns as episodes and leaves open whether they end a botnet, since several came back after weeks or years.
the other reading: takedown ends the botnet (Rustock list: dismantled March 2011; Mirai operators arrested); takedown interrupts it (Emotet resurfaced after 2021; a third Kelihos variant ran on; Torpig regained control after ten days) - C-1605 What was the first botnet?
booked: The map books the first acknowledged one in 2001 and keeps the earlier chat-room bots and attack tools as possible starts.
the other reading: MaXiTE, the first known popular IRC controller script (2002 in the list); earlier IRC bots and DDoS tools (Stacheldraht pictured as a client-server botnet; date not in the sources read); the first botnet built from devices whose owners did not know
Partners and parts published science
Built from / builds (11)
- Computer
- Malware
- Threat actor
- Command and control (C&C)
- Worker bot / proxy bot / master server (Storm)
- Denial-of-service attack
- Mirai variants for vulnerable IoT hosts
- Competition for vulnerable IoT devices among botnets
- Botnets against botnets for the same devices
- Botmasters against defenders for control of the bots
- A botnet's flood against a target's capacity
Sits inside / holds (2)
Trades with (valence) (15)
- Human (Homo sapiens) via sys:human · Botnet gain
- Internet via sys:internet · Botnet gain
- Money via sys:money · Botnet gain
- Computer via sys:computer · Botnet gain
- Domain Name System via sys:domain-name-system · Botnet mixed
- Police via sys:police · Botnet loss
- Malware via sys:computer · Botnet gain
- Intrusion detection system via sys:computer-network · Botnet loss
- Threat actor via sys:internet · Botnet gain
- Organization via sys:internet · Botnet gain
- Computer network via sys:electromagnetic-field · Botnet gain
- Denial-of-service attack via sys:internet · Botnet gain
- Cyberwarfare via sys:internet · Botnet gain
- Cybercrime and its underground economy via sys:cybercrime · Botnet gain
- Cyber threat intelligence via sys:written-record · Botnet loss
expressed by (22)
- Templating / Standardization
- Amplification/Scale-Up (Manufacture & Distribution)
- Evolvability/Plasticity Engineering (Design for Future Variation)
- Distributed Autonomy/Decentralized Execution (Subsidiarity)
- Internal Coordination/Regulation
- Sensing/Scanning
- Search/Foraging
- Navigation/Pathfinding
- Evade Detection
- Overcome Barrier
- Resource Denial/Attrition
- Subversion & Co-optation
- Prepositioning & Latency
- Leverage Host/Amplification
- Reactive Response
- Physical/Virtual Barrier
- Technique-Based Identification (a.k.a. Technique-Based Detection)
- Self-Monitoring
- Validation of Identity/Intention
- Deception
- Resilience & Repair
- Redundancy & Diversity
floor filled by (11)
- Malware
- Human (Homo sapiens)
- Internet
- Money
- Computer
- Operating system
- Domain Name System
- Threat actor
- Command and control (C&C)
- Denial-of-service attack
- Cybercrime and its underground economy
Kinds (12)
- Malware
- IRC botnet
- Peer-to-peer botnet
- DGA / domain-flux botnet
- IoT botnet (Mirai, BASHLITE, Carna)
- Smartphone botnet
- Spam botnet (Storm, Rustock, Srizbi, Cutwail)
- DDoS botnet / booter
- Data-theft botnet (Torpig, Zeus, Emotet)
- Residential proxy network
- Volunteer botnet (LOIC)
- Great Cannon (backbone-injected ephemeral botnet)
wired (3)
Notes published science
- Spam's toll: Storm sent almost 350 million pharmacy spam mails over 26 days and got 28 sales, a conversion rate under one in ten million; the researchers measured it by quietly taking over part of Storm's own command system and pointing its spam at their copy of the shop. (src:kanich-2008-spamalytics)
- Count by IP address and you get the wrong size: when UCSB took over Torpig for ten days, 1,247,642 IP addresses called in, but they belonged to only 182,800 bots. That's an overcount by an order of magnitude. The bots even turned up almost a week early. (src:stone-gross-2009-torpig)
- Mirai built one of the biggest floods on record, over 600 Gbps, from the Internet's weakest machines: cameras and routers still using factory passwords, from a list of just 62. It doubled every 76 minutes and killed rival malware on each device it took. (src:antonakakis-2017-mirai)
- Crime pays far less than it costs: a spam botnet that earned its owners about $3 million a year was costing everyone else about a hundred times that, and about half of all property crime, by volume and by value, now happens online (src:anderson-2019-cost).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
the boardroom (shared floors)
- L4: Command-and-control infrastructure: the rendezvous where bots and controller meet (IRC channels, web domains, DGA and fast-flux domains, P2P peer lists, report and loader servers, proxy bots)
- L5: The bot herder's control: commands issued, configurations and new binaries pushed, logs of active bots; signed commands in P2P botnets; master servers in Storm
- L6: Uses and the market: DDoS, spam, credential theft, click and ad fraud, mining, residential proxies; botnets rented as commodities; operators compete for size
- L7: The operator's purpose: profit, a protection racket, state power, protest (carried by the operators: threat actor's motive)
the botnet as botnet researchers describe it: a population of compromised devices up to the operator's purpose (reader's view)
- L1: The device population: Internet-connected computers, phones and IoT devices (IP cameras, home routers) with flaws and weak or default logins
- L2: Recruitment: scanning and brute force, drive-by downloads, trojans, spam lures
- L3: The bot on the host: runs, hides (MBR rootkit, renamed process), harvests, kills rivals
- L4: Rendezvous: the C&C channel (IRC, HTTP domains, DGA, fast-flux, P2P)
- L5: Control: the herder's commands, updates and counts
- L6: Uses and market: DDoS, spam, theft, fraud, proxies; rented as a commodity
- L7: The operator's purpose: profit, racket, state power, protest
way in: Check-in: the bots report themselves
- L1: The bot population, each bot on its own host, behind NATs and changing IPs
- L2: Each bot contacts the C&C every 20 minutes with a header naming itself (nid, os, cn, bld, ver)
- L3: Submissions carry harvested data and status; workers report success or error codes, relayed by proxies
- L4: The C&C keeps logs of how many bots are active and online
way in: Code from the forums
- L1: Hacker forums where code is posted (Mirai's source on hackforums.net, 30 September 2016)
- L2: Code taken up by new operators; credential lists reused (Mirai's 62 pairs subsumed BASHLITE's)
- L3: Variants built: new exploits, detection avoidance, DNS resilience (variants targeted 11 additional protocols)
way in: Customers: orders for attacks and spam
- L1: Customers who rent botnets as commodities (booter/stresser services)
- L2: DDoS-for-hire services take customers' targets (gaming platforms among them)
way in: Defenders acting on the control channel
- L1: Domains registered by researchers, servers seized, IRC channels shut, DNS entries null-routed, peer lists poisoned
- L2: C&C rendezvous unreachable, or answered by a sinkhole that returns a valid C&C response
- L3: Response: switch to the next server or domain; Zeus P2P blacklists sinkholes that talk too aggressively; Sality weighs peer reputation
way in: Harvest: what the user does on the infected host
- L1: The user's activity on the host: logins, stored passwords, web forms
- L2: Injected DLLs inspect all data handled by browsers, FTP, mail and chat clients and cmd.exe
- L3: Man-in-the-browser phishing on trigger pages set by an injection server draws out data passive watching would miss
way in: Lures and exploits: people at their machines
- L1: People at their computers: email attachments opened, web pages visited
- L2: Drive-by download, browser exploit, or a trojan run from an attachment; spam lures (a fake e-postcard)
- L3: Installer runs; Mebroot replaces the Master Boot Record and loads before the operating system; the bot calls home
way in: Peers: the P2P overlay
- L1: Other infected machines, reachable at IP addresses; the overlay's distributed hash table
- L2: Probe random IPs until another bot answers; exchange version and lists of known bots
- L3: Newer version passed on by file transfer; peer lists kept (with reputation in Sality)
way in: Police and courts
- L1: Police, prosecutors and courts in several countries (Europol, Eurojust)
- L2: Coordinated action: servers seized, domains taken over, arrests (Emotet 2021; Endgame 2024-2025)
- L3: Infrastructure taken from the operators; investigators control it
way in: Rivals on the same device
- L1: The same device held or sought by other botnets (other Mirai variants, .anime, Qbot)
- L2: Processes bound to TCP/22 or TCP/23 and processes of competing infections found
- L3: Killed
way in: Scanning: the address space and the devices that answer
- L1: The Internet's address space with its devices: IP cameras, home routers, DVRs, with default or weak logins
- L2: Stateless TCP SYN probes to pseudorandom IPv4 addresses on Telnet ports 23 and 2323, minus a hard-coded blacklist; replies show an open port
- L3: Brute-force login: 10 pairs drawn from 62 credentials; first success reported
way in: The device's own life: reboot, patch, new address
- L1: The host device restarts, is patched by its maker or ISP, or gets a new IP address
- L2: Mirai does not persist across reboot; a patch needs a reboot to take effect
- L3: Population falls (Mirai steady state 300,000 to 100,000); IPs change faster in some countries (DHCP churn factor 1.75 to 13.35)
way out: Distributed denial of service
- L1: The target's bandwidth and servers overloaded (Krebs over 600 Gbps; OVH, Dyn, Lonestar Cell)
- L3: Each bot floods (SYN, UDP, HTTP requests), or sends one request every 10 minutes to slip under volume detectors
- L4: Relayed to every bot through the C&C
- L5: Attack command issued (IRC topic "DDoS www.victim.com"; 15,194 Mirai attack commands in five months)
way out: Residential proxies
- L1: The target sees an organic internet user
- L3: Traffic relayed through the host's home connection
- L6: Proxy access sold; rotating through a pool
way out: Self-spreading
- L1: New devices taken
- L3: Bots scan and brute-force new devices; a separate loader infects them
- L5: Self-spreading set by the controller's instructions
way out: Spam
- L1: Inboxes and readers: about 350 million mails, 28 sales
- L3: Worker bots send the mail
- L4: Proxy bots relay between workers and masters
- L5: Master servers hand templates and address lists to workers
- L6: A campaign: pharmacy ads, or self-propagation under a fake e-postcard
way out: Stolen data sold or used
- L1: Accounts at banks and services (8,310 accounts at 410 institutions in ten days)
- L4: Harvested data uploaded to the C&C (almost 70 GB in ten days of Torpig)
- L6: Credentials sold online or used; credential stuffing
way out: Updates and new binaries
- L3: The bot program replaced: new domain algorithm, new exploits
- L4: Spread through C&C or peer-to-peer file transfer
- L5: Controllers distribute a new binary or configuration