Transcript Efficient utilization of communication resources for crises

Efficient utilization of communication
resources for crises management via
introducing Quality of Services (QoS)
of network traffic
Aleksandar Tsankov – Institute for Parallel Processing
Bulgarian Academy of Sciences
This work is supported by NATO Science Committee
under the project SfP 981149
NATO Advanced Research Workshop – Velingrad, October 21-25.2006
Computer networks for crises
management
Computer networks for crises management are complex
systems which consist of stationary and mobile
management centers, exchanging information via different
communication channels.
In order to ensure the efficient work of crisis management
staffs and rescue teams, crises management networks
have to have ability for sharing its resources differentially
between various users and applications according to prior
defined criteria.
That means supporting with high degree Quality of
Services (QoS) of network traffic for different users and
applications they use in process of crises management.
Computer networks for crises
management
Hierarchycal layers of computer
networks for crises management
Hierarchycal layers of computer
networks for crises management
Due to the fact that in transport subsystem are located
main part of specific functions for network management in
many cases it is identifyed with term "computer network".
In this paper are discused methods for enhancement of
transport subsystem, so when computer network for crises
management are concerned, it is mean transport
subsystems of such networks.
Ideal vs. real network for crises
managemanet
Ideal network:
no packet losses;
traffic delays are formed only from time for packet
management in network equipment;
there is no jitter in delays;
network bandwidth is suficcient for all requirements.
Real network:
network bandwidth is limited and is not enough for all
applications and users at the same time;
there are delays due to the network overloads;
there are packet losses.
Solution is to implement the Quality of Services (QoS) for
bandwidth management
Parameters of QoS
Quality of Services: giving the applications and users
predictable servicing in data delivery (transport servicing).
All parameters of network traffic which should be satisfied
by QoS fall into one of following categories:
ability for predictable bandwidth management;
minimizing packet loss and errors;
managing traffic delays and jitter.

Predictability of network traffic
Two main classes of network applications:
applications which traffic is constant flow (stream) –
characterized with high level of predictability of
generated traffic which enters the network with Constant
Bit Rate (CBR);
applications which traffic is bursty flow – characterized
with high degree of unpredictability of generated traffic
which enters the network with Variable Bit Rate (VBR);

Sensibility to delays and jitter
Asynhronous applications – no restrictions in traffic delays
(“elastic” traffic);
Synhronous applications – sensible for packet delays;
Interactive applications – delays could be noted, but not
affect functionality of application;
Izohronous applications – have a threshold of permissible
delays;
High sensible applications – packet delays disrupt
functionality of the applications.

Sensibility to packet loss
Sensible applications – practically all applications which
send/receive symbol information;
Insensible applications – many application which
send/receive data for inertial physic processes; multimedia
applications
Models for QoS management
Computer netwrok for crises management is a distributed
media which consists of many network devices. This
determines the great complexity in imposing unified
requirements fof QoS from end to end. For solving that
problem in network the management of QoS is needed.
Base architecture of QoS management includes elements
from 3 main types:
- Resources of network node – for processing the network
traffic in accordance to required QoS degree;
- Protocols for QoS signalisation – for end-to-end
coordination the work of all network devices;
- Centralized functions for QoS policing, management and
accounting.
Models for QoS management
QoS “end-to-end”
QoS node
Queues
QoS node
Conditioning
QoS node
Policy
Protocol for
QoS - signaling
QoS resources of network node
Main executive mechanisms for QoS management.
Consist of following elements:
mechanisms for queuing management – vital element
for any network equipment based on packet switching
technology. There are numerous algorhytms for queuing
management (FIFO, Priority Queuing - PQ, Weighted
Fair Queuing – WFQ etc.) ;
mechanisms for traffic conditioning – implemented in
network nodes for QoS and solving the task for creating
the needed environment for traffic servicing via
classifying, policing and shaping.
Prerequisites for implementing QoS in
computer networks for crises
management
great number of users distributed in different
geografically detached centers for crises management;
communication channels which connect centers are with
fixed limited bandwidth;
existance of great number of information flows with
different requirements for parameters of communication
channels (bandwidth, delays, packet loss);
abilities for assigning priorities and rate limits per user
basis (fine granularity/microflows);
abilities for assigning priorities and rate limits per type of
application basis (coarse granularity/aggregated flows).

Existing models for QoS management
Link-Sharing and Resource Management Models for Packet
Networks, Sally Floyd and Van Jacobson, 1995
In this paper authors introduced a solution for QoS managemet
in networks with similar requirements.
Features of S. Floyd and V. Jacobson model:
could be implemented using hierarchical queuing classes over
single network interface;
existance of various criteria for classifying, policing and
shaping the network traffic:
each interior of leaf class should receive roughly its allocated
link-sharing bandwidth
if all leaf and interior classes with sufficient demand have
received at least their allocated link-sharing bandwidth, the
distribution of any “excess” bandwidth should not be arbitrary,
but should follow some set of reasonable guidelines.
Link-Sharing and Resource Management
Models for Packet Network
Proposed solution
Things that could be futher implemented in the existing
method are:
abilities for imposing priorities and clear rate limits for
traffic consumption of each user no matter what type of
application it is generated on;
abilities for prioretizing and link-sharing the network
traffic over different types of applications;
link-sharing between user's classes depending on
current state of network consumption.
Proposed solution
Approaches for building proposed
method
1. With two routers
Approaches for building proposed
method
Advantages of method with two routers:
easily conforms within requirements;
clear and understandable way for implementation.
Shortcommings of method with two routers:
needs separate network interfaces for 2 types of traffic
management – applications and users based;
additional network equipment is needed;
in the case of already established communication
centers it could be impossible to add another network
node without changing network settings.
Approaches for building proposed
method
2. One router with support of InterMediate Queuing (IMQ)
This solution overcomes already stated shortcommings. It
is based on following elements:
Linux-based router with support of various algorhytms
for QoS;
modified kernel with InterMediate Queuing (IMQ)
support;
modified iptables package with InterMediate Queuing
(IMQ) support.
Packet traversal in the modified Linux
kernel
Abilities of IMQ
InterMediate Quqeuing device (IMQ) is an artificial network
interface existing only in Linux kernel space. It has,
however, abilities for attaching queuing disciplines on it
just like a real network interface.
This fact gives us an opportunity to implement the method
for two stage QoS on one router.
Network diagram of Two Stage QoS
using IMQ
Testing proposed model in different
scenarious
1. Initial testbed for proposed model was Computer Aided
Exercise EU TACOM SEE 2006 which took place in 23 –
24.07.2006 in IPP – BAS, Sofia. During the exercise 40
workstations and servers was working and generate
network traffic which had to be managed accroding to prior
defined criteria.
2. Two class C networks with different allocated
bandwidths and number of simultaneously working users.
Computer Aided Exercise EU TACOM
SEE 2006
Setup:
Linux-based router with IMQ support;
Backbone channel with bandwidth 10 Mb/s;
35 workstations, 5 servers, 1 DVR;
per-user rate limits of 1 Mb/s;
per-application rate limits as follows:
priority 1 class – bandwidth up to 2 Mb/s – multimedia
traffic (Skype, VoIP);
priority 2 class – bandwidth up to 3 Mb/s – SMTP traffic;
priority 3 class – bandwidth up to 5 Mb/s – WWW traffic;
priority 4 class – bandwidth up to 8 Mb/s – other traffic.

First class C network
Setup:
Linux-based router with IMQ support;
Backbone channel with bandwidth 2.5 Mb/s;
230 workstations, 3 servers;
per-user rate limits as follow:
priority 1 class – bandwidth up to 512 Kb/s;
priority 2 class – bandwidth up to 380 Kb/s;
priority 3 class – bandwidth up to 256 Kb/s;
priority 4 class – bandwidth up to 128 Kb/s.
per-application rate limits as follows:
priority 1 class – bandwidth up to 380 Kb/s – multimedia traffic
(Skype, VoIP), interactive sessions (SSH, Telnet);
priority 2 class – bandwidth up to 1 Mb/s – WWW, SMTP, POP3
traffic;
priority 3 class – bandwidth up to 1.5 Mb/s – other traffic;
priority 4 class – bandwidth up to 2.5 Mb/s – massive downloads
traffic.
Second class C network
Setup:
Linux-based router with IMQ support;
Backbone channel with bandwidth 1 Mb/s;
150 workstations, 3 servers;
per-user rate limits as follow:
priority 1 class – bandwidth up to 512 Kb/s;
priority 2 class – bandwidth up to 380 Kb/s;
priority 3 class – bandwidth up to 256 Kb/s;
priority 4 class – bandwidth up to 128 Kb/s.
per-application rate limits as follows:
priority 1 class – bandwidth up to 256 Kb/s – multimedia traffic
(Skype, VoIP), interactive sessions (SSH, Telnet);
priority 2 class – bandwidth up to 512 Kb/s – WWW, SMTP, POP3
traffic;
priority 3 class – bandwidth up to 512 Kb/s – other traffic;
priority 4 class – bandwidth up to 1 Mb/s – massive downloads
traffic.
Evaluation of the results from three
scenarious
Scenario 1 – CAX EU TACOM SEE: Due to the fact that
bandwidth of backbone channel (10 Mb/s) is sufficient for
all types of applications and user requirements, it can be
stated that tested network is close to “ideal” network (there
are no packet losses, delays and limiting one type of traffic
in favour of other). In this situation mechanisms for traffic
prioretizing are not turned on.
Computer Aided Exercise EU TACOM
SEE 2006 - results
Computer Aided Exercise EU TACOM
SEE 2006 - results
Evaluation of the results from three
scenarious
Scenario2: First class C network: Analyzing the graphics
of traffic distribution it it obvious that there are periods of
congestion in the network but they are very short and
could be neglected. It can be stated that conditions in this
network are optimal.
First class C network – distribution of
applications traffic
Evaluation of the results from three
scenarious
Scenario3: Second class C network: Analyzing the
graphics of traffic distribution it it obvious that there are
prolonged periods of traffic congestion. In these situations
mechanisms for traffic prioretizing are turned on and traffic
from high priority classes are favoured at the expense of
others. On the slide with traffic delays was shown that
there are significant traffic delays for traffic of lowest
priority class in the periods of congestion.
Second class C network – distribution
of applications traffic
Traffic delays for different classes of
applications
Conclusions
The best conditions for network users are in situation
where computer network is close to “ideal”. Such
conditions, however, are rare that's why regulation of
network traffic is a must.
Results from the study show that proposed method for two
stage management of network traffic is an effective
solution of the formulated problems.
A further goal that could be done is a thorough research of
interactions between two stages of network management
and formulation of dependencies which will allow effective
reconfiguration of network parameters in cases of arising
the changes in network environment.