and the differences in the time granularity and
run-time schedulers of JRE running on different
platforms and OS (NT, Linux, Unix). Java is not
for real time applications.
Studies of the constraints of a single GenSyn
process on a single machine show that
- The number of transmitted UDP packets per
second is limited by the processor capacity
and the memory size; - The TCP throughput is constrained by the
interface card in the sense that the TCP win-
dow mechanism reduces the throughput due to
congestion (the offered load is greater than the
capacity) before the handling of multi-threads
becomes a problem.
5.2.3 GenSyn Measurements
In the current implementation some measure-
ment functionality exists. This includes
- Source and destination ports are added to the
UDP packets generated to enable filtering
packets by tcpdump; - Trace file with records of the size of a web
page or FTP file, or the length of a phone con-
nection or a video stream; - Summary report on the total amount of sub-
mitted and received data (in bytes and pack-
ets), and the number of unsuccessful attempts.
The results in Section 2.3.1 are based on the lat-
ter summary report from GenSyn.
In the current version of GenSyn, no end-to-end
measurements of real-time performance like
delay, jitter (delay variation) and loss are done
by GenSyn. These measurements are carried out
by the use of trace software (e.g. tcpdump) on
dedicated or separate machines. Extension of the
built-in measurement functionality in GenSyn
is not realistic because of the time granularity
problems of Java. If GenSyn needs to process
each packet, e.g. add time stamp, sequence num-
ber, change TOS bit, this will significantly
reduce the performance of the traffic generator.
The solution is to make a platform dependent
function (in hardware or at least OS dependent)
that processes each packet. However, this is in
conflict with the philosophy of GenSyn that has
portability as a major requirement.
5.2.4 GenSyn Deployment
The GenSyn has been applied for the testing of
stability under establishment of an IP network.
GenSyn is also an important component in a
large-scale testbed consisting of traffic genera-
tors and external measurement machines on an
IP based test network. The testbed is prepared
and used for testing of various QoS design and
end-to-end performance of real-time applications
like voice over IP and video and TV distribution.
GenSyn has also been used in combination with
embedded load generation and measurement
equipment like SmartBits [SBit] where GenSyn
provides the background load of controllable
and realistic elastic load (TCP connections).
5.2.5 Ongoing and Planned Work
Currently, a lot of work is being done on Gen-
Syn and more is planned in the near future. The
key issues are:
Extend the model template library– In the cur-
rent version of the generator there are interface
modules to support the download of web pages
and files through http, video streaming, and
VoIP and constant packet rate. This library will
constantly be extended as new requirements
appear. In the licence agreement that accompa-
nies the GenSyn distribution, the licensee is
invited to return to the distributor all models that
are developed using the GenSyn framework.
Validation and verification– The correctness of
GenSyn models relative to the models defined is
verified, see [HeLu99]. The traffic stream from
the models defined in the GenSyn framework is
studied and one example was given in this paper.
More work on checking the validity of the mod-
els and develop new models needs to be done.
Network measurements– GenSyn is now being
deployed in a fully equipped IP platform with
DiffServ and MPLS functionality and several
different applications. The measurements from
these experiments will demonstrate the applica-
bility with respect to generating realistic traffic,
and will serve as a verification of the GenSyn
process.
The GenSyn is a Java process that generates IP
traffic using a flexible, scalable, stochastic mod-
elling framework for describing user behaviour
of sources. This stochastic behaviour model is
linked to the underlying protocol stack and gen-
erates real packets into the network. This is a
novel modelling approach that chooses the better
of two world, flexibility and scalability of com-
posite state models, and accuracy in the protocol
behaviour by use of the underlying protocol
stack instead of making a model of it.
References
[And01] Andreassen, T R. 2001. Controlled
generation of Internet Traffic.Trondheim, Nor-
wegian University of Science and Technology
(NTNU), Dept. of Telematics.