This deliverable collects the studies about the traffic handling mechanisms in the AQUILA
architecture. Some of these studies were also helpful to improve or correct some specifications
provided in D1302, as will be mentioned in the document.

Section 1 deals with the evaluation of distributed admission control schemes in Diffserv network.
A distributed admission control scheme performs admission control only at the border
of the network (ingress and/or egress), with no exact knowledge of the interior of the network.
The approach followed by AQUILA and some variants are examined and compared
with a reference scenario where hop-by-hop admission control is performed. The performance
parameters are the admission control loss and the loss in the network due to wrong admission
control decisions.

Section 2 deals with the evaluation of the mechanisms used in AQUILA to dynamically distribute
resources: the resource pools in the intra-domain and the BGRP quiet grafting in the
inter-domain. Both mechanisms have the goal to provide a scalable answer to the need of dynamic
resource management. The performance of this mechanism in term of signalling processing
reduction and network resource utilization are provided.

Section 3 deals with the analysis of packet level performance of the traffic classes proposed
by AQUILA in a larger scale network with respect to the AQUILA test bed. Packet level performance
measure like end-to-end delay are considered.

Section 4 deals with the analysis of BGRP Quiet Grafting mechanism in terms of reduction of
the path length of BGRP messages and consequently of overall number of signalling messages.

Section 5 focused on the topic of rate assurance for TCP flows. A traffic conditioning model
is developed with the goal to provide an assured goodput to TCP flows. The model is analysed
by means of extensive simulations.

Section 6 analyses the performance of the traffic class TCL 3 specified in AQUILA, focusing
in particular on the setting of the WRED (Weighted Random Early Discard) parameters. The
traffic class TCL 3 is used for long lived, greedy TCP flow (AQUILA PMM network service).

Section 7 considers the performances of the traffic class TCL 4. The traffic class 4 is used for
short lived TCP connection and for “bursty” TCP connections (AQUILA PMC network service).

Section 8 performs a comparison between declaration based and measurement based admission
control (resp. DBAC and MBAC) for IP Diffserv traffic. The comparison is focused on
the support of the AQUILA Premium VBR network service (i.e. AQUILA traffic class TCL
2). Scenarios with DBAC only, MBAC only and the coexistence of DBAC and MBAC are
analysed by means of simulation.

Section 9 deals with a modified admission control rule for TCL 3 traffic. It tries to better fulfil
the QoS requirements of the flows by taking into account the round trip time of the path and
by proper setting of the token bucket parameters.

Section 10 deals with dual-homing solutions that are used at different levels to increase the
resilience of IP networks. In particular the impact of these solutions on the AQUILA architecture
and their interaction with RCL mechanisms is considered.

Section 11 presents a scalability analysis of the AQUILA architecture. Starting from data
gathered in the AQUILA RCL performance testbed, a performance model is derived and it is
used to evaluate the scalability of the system.