Tarpit

This server's address used to belong to someone else.

Thousands of apps around the world still try to reach the server that used to live here. Nothing answers them now. We hold each connection open, say nothing, and count.

Loading…

If this is your server

Nothing is wrong and nothing was broken into. Cloud providers reuse addresses, and this one came with old visitors. They ask for websites this server doesn't host, so they never reach your mail, which keeps working normally. Holding them costs very little.

What the visitors are

Almost all of them are the same kind of app: a program, not a person at a browser, sending encrypted data dressed up as ordinary uploads, under the names of big download sites. That's the signature of tools people use to get around internet censorship. Their app still has this address saved as its server.

The pulse

Requests per hour over the last two weeks. The daily rhythm follows when people are awake where the apps are; a sudden flat stretch usually means their internet was cut or slowed.

Time zone

An average day

The same requests folded into one day: the average for each hour of the day, in the time zone chosen above. Peaks show when the apps' owners are online.

Fading away?

Requests per day since the first one was counted. As apps update or their owners move to working servers, this should slowly fall. This shows how long a dead address lingers.

from the web server's log, before the tarpit counted by the tarpit

Patience

How long a client waits for an answer that never comes before it hangs up (last two weeks). Sharp peaks are timeouts built into the app; each version of the app tends to have its own.

Where they come from

By country and by network (the internet provider that owns the address). Some networks are VPNs, so a country here is where the traffic leaves for the internet, not always where the person is.

Countries

    Networks

      Disguises

      Each request pretends to be something ordinary. It asks for a well-known site by name (none of them live here) and labels its upload as a video, a sound file or a spreadsheet. The first bytes of the upload tell us whether the label is true.

      Sites they ask for

        What the upload claims to be

        LabelRequestsLabel true

        "Label true" checks the file's first bytes (a real FLV video starts with FLV, a WAV file with RIFF). Blank means the label makes no checkable claim.

        Upload sizes

        Randomness of the upload

        Bits per byte in the first part of each upload. Text is around 4–5; close to 8 means encrypted or compressed. Short uploads can't score high even when random.

        Fingerprints

        How the clients identify themselves. A single program name and a handful of encryption profiles across hundreds of thousands of addresses suggest one app with many users.

        Programs (user agent)

          Encryption profile

            Protocol, negotiated app protocol, and a short hash of the ciphers and curves the client offers. Coarser than a JA3/JA4 fingerprint, which needs details the web server doesn't expose.

            HTTP version

              Other visitors

              Every public address also gets scanners: automated programs probing for known weaknesses or open proxies. They get a plain "not found" right away. Top requests in the last two weeks: