Empirical Evaluation of Hash Functions for PacketID Generation in Sampled Multipoint Measurements

Abstract

A broad spectrum of network measurement applications demand for multipoint measurements; e.g. one-way delay measurements or packets path tracing. A passive multipoint measurement technique is realized by generating a timestamp and a packet identifier (ID) for each packet traversing an observation point and sending this information to a common collector. The packet ID can be provided by using parts of the packet or generating a digest of the packet content. Multipoint measurements demand for high resource measurement infrastructure. Random packet selection techniques can reduce the resource consumption while still maintaining sufficient information about most metrics. Nevertheless random packet selection cannot be used for multipoint measurements, because the packets selection decisions on its path can differ. Hash-based selection is a deterministic passive multipoint measurement technique that emulates random selection and enables the correlation of a selected subset of packets at different measurement points. The selection decision is based on a hash value over invariant parts of the packet.

When hash-based selection is applied two hash values are generated - one on which the selection decision is based and a second one that is used as the packet ID. In a previous paper we already evaluated hash functions for hash-based selection. In this paper we analyze hash functions for packet ID generation. Other authors recommend the use of two different hash values for both operations - we show that in certain scenarios it is more efficient to use only one hash value.