MAISIE MODELS
Models Currently Available
Abstract
MIRSIM is a parallel implementation of Irsim, a public domain switch level
simulator. MIRSIM can be used to execute models, without any changes,
with a variety of simulation protocols on both distributed memory
and shared memory parallel architectures.
Circuits which will be simulated in parallel are first partitioned
among a number of subcircuits, such that the computation among the
subcircuits is approximately balanced and the communication is
minimized. We have developed an Interactive Circuit Partitioning and
Simulation Environment (ICPSE) to ease the partition process.
In ICPSE, a VLSI circuit, which is designed under
the Magic environment, can be partitioned manually, automatically, or
semi-automatically. ICPSE generates a netlist file which
is backward compatible to Irsim's format and provides a GUI for users
to start simulation on a parallel machine and examines the simulation results
graphically.
The partitioned design is subsequently simulated on a
parallel architecture by executing one or more partitions on each processor.
Synchronization among the partitions is required to ensure that incoming
signals at each subcircuit are processed in their correct global
order. Two primary synchronization mechanisms have been used: the
conservative parallel discrete-event simulation (PDES) protocols and
the optimistic PDES protocols. Six benchmarks (ranging in size from 3K
transistors to about 87K transistors) have been tested to evaluate
the performance of MIRSIM as well as the effectiveness of different
partitioning algorithms. In general, manual partitioning performs better
than automatic partitioning. However, different benchmarks favor
different synchronization protocols and the speedups measured vary among
benchmarks and protocols. We have already measured a maximum speedup of 5.8 using
an conservative protocol and 5.7 using an optimistic protocol on the IBM SP1 machine with 16 processors.
Machines
MIRSIM runs on UNIX workstations and the IBM SP.
More Information
- Parallel Switch-level Simulation of VLSI Circuits, Yu-an Chen, Vikas Jha, and Rajive Bagrodia, Computer Science Tech Report:950020, June 1995, UCLA.
- Parallel Gate-level Circuit Simulation on Shared Memory Architectures, Rajive Bagrodia, Yu-an Chen, Vikas Jha, and Nicki Sonpar, IEEE, Proceedings of the Ninth Workshop on Parallel and Distributed Simulations, pp.170-174, June 1995, Lake Placid, NY.
- Related MIRSIM web page.
- Or contact Yuan Chen at yuan@cs.ucla.edu.
Description
This channel modeling utilizes the SIRCIM impulse responder parameters
to characterize the radio propagation model, i.e. multipath, shadowing
effect, spatial correlation. In order to preserve the signal continuity, this channel model will remember the history of node location.
The channel model provides three important choices:
- free space distribution
- shadowing, log-normal distribution
- SIRCIM channel fading model
A sequential version of this model is available, and a parallel version is being developed.
More Information
For more information contact Eric Wu at hsiao@cs.ucla.edu.
Description
The WAMIS simulator is used to model the clustering-based instant
infrastructure approach designed at UCLA toward supporting multi-hop
wireless and mobile packet radio networks. The simulator simulates
various protocols at different layers in the protocol stack. The
simulator, as of now, simulates T/CDMA as the channel access method.
It also simulates various routing strategies such as DSDV and
QOS routing for a mobile and wireless environment.
The entities are described as follows:
- entity traffic_generator: This presents the application layer which
can generate user traffic, independent of the network protocols.
- entity node: The entire networking algorithm is placed here. The
possible messages include receiving_pkt (control/data),
time_to_transmit, channel_busy (if use CSMA), get_pkt_from_trfc_generator,
etc.
It could invoke pkt_message to channel entity, invoke rec_pkt to
application entity, invoke node_location to channel entity, etc.
- entity channel: All the packets transmitted should be gathered here, and
it decides which nodes can get which packets, using the propagation
model and node location information, power, etc.
A sequential version of this model is available, and a parallel version
is being developed.
Machines
The simulator runs on Solaris, SUN OS 4, and SP2.
More Information
Description
The SSN simulator models a wormhole-routing network down to the
byte-level. It models both a local-area part, which is based on
the Myrinet, and a larger-area optical part. The basic entities
are hosts, electronic switches and optical switches. These are
interconnected arbitrarily in a topology specified by a runtime
file. Other features modeled include multicasting and multiple
priority host queues. Routing can be performed automatically by
one of several algorithms at runtime, or it can be specified
explicitly to the program.
Machines
The simulator runs on a Sun SPARCstation with SunOS 4.
Availabilty
The simulator is currently available. For details contact Simon Walton at simonw@cs.ucla.edu.
More Information
Descriptions
Connectionless traffic sources cannot specify the values of the
traffic descriptors required by ATM networks for resource
reservation. For this reason, connectionless traffic will be carried on
an Available Bit Rate (ABR) basis, allowing it to fill in the
bandwidth left over by services for which resources have been
allocated, but with no performance guarantees. Congestion control
techniques have been devised for the control of ABR traffic over ATM
networks that do not interact well with the connectionless traffic
generated by TCP sources. This simulation model using MAISIE was
developed to study precisely this interaction, comparing the
effectiveness of different ABR congestion control schemes.
Our simulation models an ATM network, consisting of a number of ATM
switches connected in tandem. We use the ATOM switch architecture as a
model for the basic switching functions, and specific features are
added at ATM and AAL levels in order to support ABR congestion control
schemes like PRCA, EPRCA, SP-EPRCA and FCVC. All traffic sources share
a single outgoing link, where bandwidth is allocated only to
guaranteed sources, while best-effort sources are competing for the
residual bandwidth. A more detailed description of each element of the
simulator follows.
Guaranteed traffic generator
For the sake of simplicity we assume the guaranteed traffic comes
through a single input link. It is generated as an MMDP process
resulting from aggregating VBR sources. This traffic is used to
simulate variations on the bandwidth available to TCP sources by
varying the number of simultaneously active guaranteed sources.
Hence, we can study which scheme allows for a better fill in
characteristic. The traffic generator is implemented by a MAISIE
entity.
ATM workstations
The connectionless traffic is generated by an ON-OFF source with
exponential active and idle periods. This source provides application
messages to the TCP layer for segmentation prior to transmission on
the ATM network. All TCP connections are then kept open throughout
the simulation (20 seconds of actual time) and the TCP entity at the
source site performs window-based flow control functions. For our
purposes, the IP level performs minor functions. A 20 byte header is
added to each TCP segment.
The AAL level is based on the AAL5 protocol. AAL and ATM levels also
perform operations related to the congestion control schemes
under study. The reverse direction is assumed to be uncongested, thus
no credit cell, RM cell, or TCP acknowledgments are lost in our
simulation. We are mainly concerned with congestion inside the
network, therefore TCP segments are assumed not to be lost either at
the source or destination side.
The TCP and application message generation are implemented by the same
MAISIE entity while the AAL and ATM functions are performed by another
entity. In both cases, different entities are used at the sending and
receiving sites. It is the AAL/ATM entities on the sending and
receiving sites that perform the ABR flow control functions
needed at the edge of the ATM network.
ATOM switch
The simulation of the ATOM switch is based on three MAISIE items. The Input
Packet Processor performs ATM cell header processing and tags cells
for internal routing. The Broadcast Bus routes cells at speed N times
the link speed (up to N cells can be switched per time unit). Cells
are queued in the appropriate output buffer and scheduled for
transmission by the Output Packet Processor. Double priority buffering
is used. ABR cells are taken from the opposite buffer only when the
guaranteed cells buffer is empty. It is the Output Packet Processor
entity that performs the ABR functions required inside the ATM
network.
Machines
This simulation was run on a Sun Sparc Station 5 running Solaris.
Availability
This model is presently available for use.
More Information
-
Performance of TCP over ATM for Various ABR Control Policies,
C. Pazos, V. Signore, D. Cavendish Jr., M. Gerla,
Proceedings of ICCCN '96, Rockville, MD. 1996.
- Comparing ATM Credit-Based and Rate-Based Controls for TCP Sources, M. Gerla, C. Pazos, and V. Signore, Proceedings of MILCOM '95, San Diego, CA.
- Also contact Carlos Pazos at pazos@cs.ucla.edu or Dirceu Cavendish at dirceu@cs.ucla.edu.
Description
The mobile wireless network system is an advanced simulation environment
which is used to examine, validate,
and predict the performance of mobile wireless network systems. This simulation environment overcomes many of the limitations found with analytical models, experimentation, and other commercial network simulators available on the market today. We have identified a set of components which make up mobile wireless systems and have created a set of flexible modules which can be used to model the various components and their integration. These models are developed using the Maisie simulation language. By modeling the various components and their integration, this simulation environment is able to accurately predict the performance bottlenecks of a multimedia wireless network system being developed at UCLA, determine the trade-off point between the various bottlenecks, and provide performance measurements and validation of algorithms which are not possible through experimentation and too complex for analysis.
More Information
Description
Accurate simulations of parallel programs for large datasets can often be slow; parallel
execution has been shown to offer significant potential in reducing the execution time of many discrete-event simulators. This model is the implementation of a parallel
simulator called DPSIM that simulates the execution of data parallel programs on
contemporary message-passing parallel architectures. DPSIM has been used for a variety of applications, including Gauss Jordan elimination, fast Fourier transforms, and matrix multiplication.
Machines
The simulator has been implemented on the IBM SPx using a conservative synchronization algorithm.
More Information
Description
ATN is a UCLA MICRO Teledyne project. The maisie model was structured from the ground up with the messaging structure and the conditional receives in mind. The model is broken into functional entities. The airplane and groundstations had common structure which encompassed the following processes: Receive, MAC, Transmit and LLC. The ground station further had the Route process while the Airplane had an arrival and transport process. The airplane has packet generator process and terminates the received packets.
The Arrival process at the airplanes generate packets according to the distribution specified. The packet is then processed by the OSI layers and sent out to the ground station in control of the aircraft. The MAC layer specifies this access mechanism. Currently, CSMA, CDPD and DRMA protocols have been implemented. The ground station has a route
process which is responsible of finding the ground station in control of the destination aircraft of the packet. This routing is assumed to be carried out by the back bone network. In the simulation, the nodes are assumed to be connected directly.
Machines
The current version supports parallel runs on a network of workstations, SP2, a connection machine or a Sparc 10 Server.
More Information
Description
The Exponentially Correlated Random (ECR) mobility model simulates the
movement of nodes in a multihop packet radio network in a tactical
setting. The model can have several groups of nodes. Each group as a whole
moves according to the model, and each node within a group also moves
according to the model, but following the trajectory of the group. The
idea is to simulate the movements of multiple networked military units
(e.g., brigades).
The model is controlled by the equation
b(t+1) = b(t)exp(-1/Tau) + sigma * sqrt(1 - (exp(-1/Tau))^2) * r
where:
- b(t) is the position (r, theta) of a group or a node
- Tau is a time constant (regulates the rate of change from 1 time
step to the next)
- Sigma is the variance (regulates range of change)
- exp is the transcendental number e (~2.7183)
- r is a random gaussian variable
Each group is a circle containing nodes initially spread out from the center
with a radius R.
Each group's motion is described by moving a certain radius with a certain
angle. A set of Tau and sigma variables are specified for the radius and
a second set is specified for the angle of the group. In general, the
smaller the Tau, the more random the movement. Sigma is used to control
what the spread is of the speed or the angle movement is.
Each node's movement is specified just like a group's. Specifically, all
nodes within a group are given a set of Tau and Sigma variables for the
radius and angle. Nodes in different groups may have different sets
of variables.
Machines
The model runs on Sun SPARCS with SunOS 4.1.x.
Availability
The model is currently available by contacting Regina Rosales Hain
(rrosales@bbn.com), BBN Corporation.
Description
This model simulates a generic link-state routing algorithm that generates
routes using Dijkstra's shortest path algorithm in a flat (non-hierarchical)
network consisting of possibly mobile nodes. Currently, free space
propagation is assumed, that is, there exists a link between two nodes
if and only if their euclidean distance is below a threshold. Link up/down
events, caused by mobility or other models orthogonal to this model, trigger
link state updates that are flooded throughout the network. Periodic
updates are also sent. Source routes are generated on request between a
source and a destination. It does not maintain a routing table - packets
are expected to be source routed.
Machines
The model runs on Sun SPARCS with SunOS 4.1.x.
Availability
The model is currently available by contacting Regina Rosales Hain (rrosales@bbn.com), BBN Corporation.
Description
The goal of this model is to evaluate scalability issues
and reconciliation algorithms for replicated file systems.
Primary components of the model include:
- Rumor Simulation: this module simulates the detailed algorithms of Rumor file
replication and reconciliation module,
including the reconciliation topology, reconciliation of version
vectors, major disk accesses, reconciliation overhead, and end-to-end
rumor daemon hand shakes.
- File Reference Generator: the file reference generator takes in the file
access distribution and file operation distribution and
generates file accesses accordingly.
- File System Simulation: this module keeps track of file attributes.
- Disk Simulation: this module simulates the disk delays, including seek time,
rotational latency, and transfer time.
- Network Topology Simulation: this module will support different physical
topologies.
Availability
Expected availability: August '96.
More Information
For more information contact Andy Wang at awang@cs.ucla.edu.
This web page developed by Monnica Terwilliger. Bad link? Send e-mail to
monnica@cs.ucla.edu.
Last modified: February 25, 1998.