ZOO
A
tool for traffic analysis and characterization
Zoo is a tool for traffic characterization and
analysis that has been developed in the framework of the French Metropolis project. The following
gives motivations and goals of such a tool, as well as some information on
measured parameters, user manual, and so on. Zoo is presently used in the EuQoS European project. Zoo has been designed
and developed at LAAS by:
-
Philippe Owezarski (project leader)
-
Nicolas
Larrieu
- Hubert Martin-Deidier
Zoo belongs to CNRS-LAAS. However, for
interested persons, the tool is freely available and can be downloaded at the
bottom of this page. However, LAAS does not
provide any maintenance service for this tool. Nevertheless, even if an answer
is not guaranteed, in case of problem with Zoo you can try to contact Philippe Owezarski.
Internet traffic characterization and analysis show
that Internet traffic model and characteristics are very far from common
beliefs and that it more and more exhibits complex properties as
self-similarity and long range dependence (LRD), properties that are very
damageable for traffic profile and network QoS. Modeling such traffic is then
quite difficult, and it seems that the required mathematics are not yet
existing. Therefore, it is quite impossible to evaluate the performance of
network in the presence of such complex traffic.
Zoo then tries to find a way for measuring quality of
service, finding parameters for defining the traffic process, and helping to
issue a suited model for the traffic. For achieving this, Zoo has the following
functionalities:
·
It makes possible to describe the traffic
process by computing statistical moments of the traffic. In theory, all moments
from first to infinite orders are required. Practically, first (mean) and
second orders (standard deviation, auto-covariance) are sufficient to have a
good representation of traffic process characteristics.
·
Zoo measures the quality of service provided by
the network in terms of dynamic traffic parameters. In a paper
published at HSNMC’2004, Zoo’s authors showed that LRD is a good way to
qualify and quantify the high traffic variability. The principle of this QoS
evaluation is then based on the oscillation nature, LRD and self-similarity
levels (that are quite close notions for Internet traffic), given that the more
oscillations and LRD, the lowest the QoS. Therefore, Zoo aims at studying the
traffic dynamics, and especially the LRD properties of the traffic, LRD
being one of the main parameter for qualifying and quantifying QoS. Note
that LRD is evaluated thanks to the LDestimate
tool from Patrice Abry and Darryl Veitch.
·
Properties as LRD or self-similarity are
related to the high variability of the traffic. Analysis results of this high
variability phenomenon exhibited that it is due to the transmission of long
flows consisting of a large number of packets - called elephants - using TCP.
Short flows (mice) and elephants then have different characteristics when
considering their related traffic. Similarly, traffic related to different
applications has different properties. Therefore, for helping traffic analysis
and modeling, Zoo allows users to analyze the characteristics of different
classes of traffic. These classes can be defined as users want: as concrete
applications, depending on the size, or duration of flows, etc. This can
provide very important information for handling these different traffic classes
differently, and then adapt the protocol and mechanisms to their real
requirements.
Measured parameters:
static:
-
number of packets and bytes in the trace (for
TCP or UDP, or globally)
-
number of flows (for TCP or UDP, or globally)
-
number of packets, bytes and flows according to
traffic decomposition into applications
-
number of TCP acks
-
number of TCP losses
dynamic:
-
Evolution of global throughput during time
-
Evolution of TCP and UDP throughput during time
statistic:
-
inter-arrival time distribution for TCP flows
-
flow length distribution for TCP flows
-
flow duration distribution for TCP flows
-
inter-arrival time distribution for TCP packets
-
TCP packet length distribution
Because of the algorithm complexity and the size of
traces analyzed, Zoo has to work on a powerful computer with a large memory. A standard
hardware configuration is Pentium IV at 2 Ghz with 1 Go of RAM. This software
was developed under Linux operating System and needs a Kernel 2.2. Moreover,
graphical user interface (GUI) is based on GTK 2
(Gimp Tool Kit). Thus, GTK library must be installed otherwise Zoo will not
work well.
Note that Zoo works with traffic traces captured by DAG cards from ENDACE.
However, Dagconvert, a tool for converting
traces from the classical tcpdump format to the DAG format is provided in the
download part of this page.
To get the user manual, just click here.
For the moment none.