Everything the sensors have recorded

What you would be buying, counted. Our sensors answer as exposed servers across the internet and keep what unsolicited traffic does to them: the request, the body, the file it tried to drop, the credentials it sprayed and the fingerprint of the client that sent it.

Breakdowns measured 27 October 2026 over 30 complete days. The headline figures are live.

The ports that get knocked on

Ports by traffic over 30 days, with the number of distinct addresses behind it. The two counts disagree on purpose: one loud host moves the probes column and not the addresses column.

PortWhat answersProbesAddresses
39131unassigned, swept as a pair with 39132893,9073,847
443HTTPS246,5374,889
80HTTP232,9773,937
22SSH188,5823,294
PortWhat answersProbesAddresses
445SMB134,8417,939
1433Microsoft SQL Server77,7333,827
3unassigned, low-port sweep75,465241
10unassigned, low-port sweep74,487245

The protocols on the wire

Application protocols recognised over 30 days, by probes and by the addresses that spoke them. SSH and RDP carry fewer addresses than SMB and far more attempts each.

ProtocolProbesAddresses
HTTP4,328,70337,967
SSH957,8326,197
BitTorrent895,6054,086
RDP589,9534,178
ProtocolProbesAddresses
HTTP proxy543,792891
TLS147,3325,825
SOCKS595,5241,162
MSSQL TDS85,4954,968

The countries it comes from

By distinct source addresses over 30 days, which is the ordering that holds: ranking by probes would put one busy network above a country sending thousands of separate hosts.

CountryAddressesProbes
United States20,1491,947,378
China7,413583,841
Belgium3,45739,848
United Kingdom2,513489,426
CountryAddressesProbes
Germany2,3421,620,599
India1,74792,691
Singapore1,37059,344
The Netherlands1,2071,108,209

The method behind these numbers

Stated so you can check them, and so the ones that cannot be checked are visible as such.

The window

Thirty complete days. Both edges of a bucketed series are unsettled, so a partial first or last day is dropped rather than drawn.

The population

Unsolicited traffic reaching sensors that answer as exposed servers. It is not a sample of the internet, and it is not traffic anyone asked for.

Addresses, not people

A distinct address overcounts one operator rotating hosts and undercounts a network behind one address. Treat it as an ordering, not a census.

Two columns on purpose

Probes and addresses are reported side by side everywhere, because one loud host moves the first and not the second.

Nothing about the sensors

Not how many there are, not where they sit, not when any of them started. That is deliberate.

Check any of it

Every figure has a page behind it, and ?format=json on that page returns the same numbers as a document you can diff.

Query it before you decide anything

A free key takes a minute, needs no card and runs 500 queries a day. The public feeds need no account at all.