120 likes | 128 Views
This report provides an update on the PSAMP information model for packet sampling, including changes from the previous version and open issues that need to be addressed. It also discusses sampling and filtering functions, match filtering methods, method chaining, and the observation point.
E N D
PSAMP Information Model Status Information Model for Packet Sampling A Status Report Thomas Dietz dietz@netlab.nec.de Falko Dressler dressler@informatik.uni-tuebingen.de
Changes from Previous Version Restructerd field order Added parameter probability for uniform probabilistic sampling
Open Issues Fields for some sampling methods and almost all filtering methods need to be defined The chaining of filter and sample methods must be defined Usage of IPFIX fields in PSAMP must be described Number space for field IDs (currently starting from 1024)
Sampling/Filtering Functions Sampling and filtering methods may contain functions Non-uniform probabilistic sampling Flow state sampling Hash filtering Router state filtering Do we need standard functions for those methods?
Match Filtering (1) Match filtering is currently defined as a bit mask on the first 20/40 bytes of a IPv4/IPv6 packet Advantages Only one field is needed to encode a match Very flexible for future extensions (no changes to info model needed)
Match Filtering (2) Disadvantages Option/Extension header fields cannot be matched Transport protocol header fields are difficult to match (impossible if there are extension headers present) Difficult to read and encode Difficult to implement
Match Filtering (3) Proposal: Create a simple description language Header fields can be matched easily Most existing devices have a filtering language anyway (access lists) If extensions are needed they can be defined by vender specific fields (in a first step and can be standardized later if needed)
Method Chaining (1) Several sampling/Filtering methods can be applied one after another How do we represent this feature in the info model? Proposal: Several fields containing a template ID One field containing all template ids
Method Chaining (2) – Proposal 1 Several fields containing a template ID Template ID ... Option Data ID 1 Option Data ID 2 ... Advantages Easy to encode Disadvantages The order of the template IDs is not guarantied by the protocol Does only work if order is significant
Method Chaining (3) – Proposal 2 One field containing all template IDs Template ID ... Option Data ID List Advantages Order of the template IDs is implied by the order within the field Always one field regardless of the number of methods chained Disadvantages More difficult to encode
Observation Point The observation point is currently not defined in the info model This should be taken from the IPFIX info model
The End Thank you for your attention