High-interaction honeypots vs. low-interaction honeypots When people think of honeypots, they often think of complex, highly realistic "traps" where the hacker encounters a range of fully functional services (a realistic website, an email server with updated emails, and so on) and his every move can be tracked. These types of high-interaction honeypots provide realistic emulation of high-value network assets in return for significant administrator effort. Their sophistication is intended to better determine the hacker's motivations and to better document what the hacker did.
For example, once when I was onsite at a large defense contractor, a newly installed honeypot caught someone probing the SharePoint Web server. We quickly set up three areas of the site designed to help us profile our intruder: a section with computer games, a section hosting "secret" NASA Space Shuttle plans, and a section that purported to have F17 fighter pilot communication codes. The secret Space Shuttle plans were simply page redirects from NASA's public website. The hacker quickly went to the Space Shuttle plans and began using SharePoint's search feature to look for Middle East topics. This was no gamer. The hacker was later found to be a foreign spy working in the company's accounting department as a temp worker.
Because high-interaction honeypots require a lot of work and carry increased risk that an attacker will use the exploited honeypot to do harm (for example, attacking other companies, installing a password sniffer, and more), I encourage most companies to use low- or medium-interaction honeypots. A medium-interaction honeypot fakes common tasks, but doesn't implement a full service. For example, a fake FTP service might allow the prober to attempt to logon, or it might allow them to logon anonymously and offer up fake files to download. A fake email server might even let the attacker read and send emails. KFSensor allows a few emails to be sent on the fake service so that a potential spammer might be tricked into thinking he's found a real email server. The idea is to provide enough functionality to determine whether an intruder poses a threat, but not enough to allow the intruder to take things too far.
Low-interaction honeypots are the simplest of all. Honeypots that serve as early-warning systems are usually low interaction, meaning that they monitor one or more network ports and alert when something has tried to connect to a particular port. Low-interaction honeypots don't attempt to look like fully formed, legitimate services. Attackers rarely understand why the remote port isn't responding correctly, and they move on after a few attempts. That's OK, because you've hopefully logged the origination point of the probes and are now exploring it yourself.