DShield Principles
DShield.org's system was designed to be fast and scalable and
to include a wide range of sensors. The initial challenge was to define
the lowest data fidelity considered valuable to the system. We chose
to focus only on very basic TCP/IP header information: date and time
with one-second resolution, source and target IP and ports, protocol
and flags. This allowed DShield.org to include a number of non-IDS
packages, such as personal firewalls. While we lose some valuable
information from other systems, such as IP and TCP options and packet
content, we gain a large number of additional sensors. Also, the amount
of data that has to be processed is reduced, allowing for a more agile
system while increasing the number of sensors.
From the beginning, DShield.org rejected relying on signature-based
recognition at the client. This would have significantly complicated
integrating new IDS packages, as each one uses its own set of signatures
and its own set of labels for any of the thousand or more different
attacks these packages recognize.
We also decided not to issue any specific guidelines for users
as to what event to submit or not to submit. Any such guideline
would introduce a bias into the data and could skew it. Instead,
if asked, DShield.org recommends the submission of any event perceived
as "irregular" or "hostile" by the user.
|