View
218
Download
0
Category
Preview:
Citation preview
How may networks be studied?
Measurements from real devices / networks
Measurements from real devices
Protocol analyzers, i.e. “sniffers”
Operator conducted drive tests with specialized equipment
Networks / devices may not exist for state-of-the art wireless technologies
Test networks
Access to all equipment / software
Open-source software
Test networks may not be large enough
Computer simulations
Faster to achieve results
Easier to analyze with different assumptions
How reliable are the results?
Mathematical analysis
Pen’n’paper, Matlab, Mathematics, etc.
2 © 2012 Magister Solutions Ltd
Magister Solutions – From science into business
3 © 2012 Magister Solutions Ltd
Mobile Applications
Active
In-house mobile software
development
R&D Training Analytics
New Active Active
Educational, academic and expert networks
Universities EDU cluster
VTT Corporate
development
“Essentially, all models are wrong, but some are useful”
[George E. P. Box and Norman R. Draper. Empirical Model-Building and Response Surfaces. Wiley, 1987]
4 © 2010 Magister Solutions Ltd
When to use simulations?
When the analytical model/solution is not possible or feasible
Many times, simulation results are used to verify analytical solutions in order to make sure that the system is modeled correctly using analytical approaches.
Dynamic systems, which involve randomness and change of state with time
Complex dynamic systems, which are so complex that when analyzed theoretically will require too many simplifications. In such cases, it is not possible to study the system and analyze it analytically.
5 © 2012 Magister Solutions Ltd
What are simulations used for?
Generally
Test improvements and their gains
To identify bottlenecks in a process
To prevent under of over-utilization of resources
Provide a safe and relatively cheaper (in term of both cost and time) test bed to evaluate the side effects
Optimize the performance of the system
6 © 2012 Magister Solutions Ltd
Simulation scenarios
With wrap-around radio conditions of the actual simulation area are replicated around the actual simulation area where UEs are located
Non-regular layout
7 © 2012 Magister Solutions Ltd
R&D examples
8 © 2012 Magister Solutions Ltd
Spectrum 1 Spectrum 2
5 MHz
DC capable UE
Multipoint Tx
0 5 10 15 20 25 30 350
5
10
15
20
25
30
35
40
45
50
Users per cell
Mean S
ofter
HO
User
BR
Gain
(%
)
PA3 Channel (Softer Handover Users)
Tokyo example
Goal:
Performance benchamarking between 3G HSDPA and next generation LTE system
Challenges:
It is hard to collect needed statistics from commercial networks
It is not affordable to build large enough test networks
In relation to LTE, there was only limited commercial products available
9 © 2012 Magister Solutions Ltd
References – Case Tokyo
Simulation based approach was selected
Like many vendors, operators as well as the scientific communities do for studying the wireless cellular network performance
Digital network planning data over Tokyo map was used in the simulator
Realistic conditions through non-regular network layout and propagation
10 © 2012 Magister Solutions Ltd
Available simulators
Most suitable:
Network Simulator-2 (NS-2) http://isi.edu/nsnam/ns/
Network Simulator-3 (NS-3) http://www.nsnam.org/
OMNeT ++ http://www.omnetpp.org/
openWNS https://launchpad.net/openwns
Other open
JiST http://jist.ece.cornell.edu/
GoMoSim http://pcl.cs.ucla.edu/projects/glomosim/
Proprietary
OPNeT http://www.opnet.com/
QualNet http://www.scalable-networks.com/products/qualnet/
NetSim http://tetcos.com/software.html
Company owned
11 © 2012 Magister Solutions Ltd
Simulator selection criteria Parameter NS-2 NS-3 OMNeT++ openWNS
License GNU GPLv2 GNU GPLv2 Free for academy
and education LGPL
Wireless technologies
+ + + + -
Support/development
+ - + + + -
Real time integration
+ -
+ + -
Platform C++ and OTcl C++ and Python C++ and NED Python with
C++ libraries
Performance - + -
12 © 2012 Magister Solutions Ltd
NS 3 Organization
tests
helper
routing Internet stack
devices applications
node mobility
common simulator
core
Smart pointers, Dynamic type systems, Attributes, Callbacks,
Tracing, Logging, Random Variables
Events, Schedulers, Time arithmetic
Packets, Packet tags, Packet
headers, Pcap/Ascii file writing
Mobility models (static, random
walk, etc.)
Node class, NetDevice, Address types (Ipv4, MAC, …), Queues,
Socket, Packet sockets
13 © 2012 Magister Solutions Ltd
NS3: Main objects, attention to realism
Node class
Protocol stack
Application class
Channel class
Net device class
Topology helpers
Packets
Sockets-like API
• Model nodes like real computer • Support key interfaces such as sockets API
14 © 2012 Magister Solutions Ltd
Discrete-event simulation
The idea is to jump from one event to another
Events are recoded in a future event list (FEL)
Each event notice is composed at least to data: time, type
Event routine or handler to process each event
Each event may change the system state and generate new event notices
15 © 2010 Magister Solutions Ltd
Simulator types
Different ways of dividing simulators
Running principle: discrete event, continuous, …
Level of detail: static, quasi-static, dynamic, …
Modeling purpose: wireless radio access, wired, …
Confidentiality: proprietary, open
16 © 2012 Magister Solutions Ltd
Simulation dynamics
Moving randomly
Yes
Fully modelled RRM
Dynamic simulations
Varied every timeslot
(slot, TTI, etc.)
Mobility Study, RRM
Study, etc.
Slow
Static simulations “Snapshot”
Static
No
Simplified and limited
algorithms, e.g no outer
power control
Fixed
Algorithms
Path Loss
Time Domain
Mobility Static
Extensively modelled
RRM
Fixed
Yes
Quasi-Static Simulations
RRM study excluding
mobility, etc.
Middle
Interference study between
systems, etc. USE Cases
Fast Computational
Complexity
System simulators
17 © 2012 Magister Solutions Ltd
Link and system level simulators
(Mobile) network simulators can be roughly categorized in link level and system level simulators
Link level simulators model in high detail the radio interface between a MT and BS. Link level simulators operate on chip or symbol level
System level simulators model a full network usually including multiple BSs and number of mobile stations. Slot-level modeling is sufficient for system simulation (1 slot = 2560 chips = 0.66667 ms)
Link level simulator output works as input for system level simulators using mapping tables or curves
18 © 2012 Magister Solutions Ltd
Traffic models
Defined by standardization
organizations
Infinite buffer
FTP
Constant Bit Rate (CBR)
HTTP models (bursty)
AMR codecs for VoIP
22 © 2012 Magister Solutions Ltd
Dpc
Nd
packet callpacket call
embedded objects
(Reading Time)
main object
Packet calls
Dpc
Packets of file 1 Packets of file 2 Packets of file 3
Metrics and performance indicators
Cumulative distribution function (CDF)
Line plots
Bar plots
Map plots
Traces
23 © 2012 Magister Solutions Ltd
Generic architecture
Simulations
...other
Parameters
Results
Run/debug/visualise
encoder decoder
Player (NetAnim, PyViz,…)
Scenarios
24 © 2012 Magister Solutions Ltd
Post-processing tool
Recommended