Meetings with Hans   
   
Meeting with Prof. Hans

One-to-One Meeting on Every Friday morning from 10:00am to 11:00am



One-to-One Meeting on Friday afternoon from 15:00am to 16:00am (21 November 2008)

    Talked about:

    1.There four levels of Modelling and Simulation (From bottom to top):

      1)Machine, and sometimes machines distributed

      2)Language like C++, Python, Java

      3)Simulation Moduling method like DEVS, CONT, DISC, CELL, AGENT

      4)The modelling method like graphs in DSM.

    2.MSDL now focusing on the modelling transformation and the development of strong simulator.

      1)The language like C++ is too low to do the modelling, the combination and tracing is difficult be automatic.

      2)DEVS has good definition to do the modelling and simulation, it is more suitable to run in special simulator.

      3)The model transformation which is being done by Eugene could transfer from other Simulation formalism to DEVS.

      4)Only transformation and DEVS simulator is needed to do modelling in all kinds of modelling methods.

      5)We need a strong DEVS simulator to support the complex DEVS model, considering not only the executive ability but also the efficiency.

    3.Distributed DEVS done by Eugene and Amr.

      1)Eugene do the modelling of the simulator. This work is to simulate the working process of distributed simulator.Then the simulation result could be used to help implement the real simulator.

      2)Amr use Pydo to send message between DEVS simulators from machine to machine.

      3)The paper about distributed DEVS could be written to submit to the PEDS conference.

    4.Something about research.

      1)Learn more about the DEVS simulation.

      2)Be plugged into model transformation.

      3)Focus on the distributed DEVS simulation, may be use HLA, or others.



One-to-One Meeting on Wednesday afternoon from 15:30am to 16:30am (3 December 2008)

    Talked about:

    1.Introduction of Time Warp in detail:

      1)The modules run on different LPs advance to different logical time at the same time on the wall.

      2)The messages with timestamp should be sent and received according to the time sequence.

      3)So when the straggler comes, the module should do the ROOLBACK, and send the Anti-Messages to the modules which accept the messages before the time straggler came.

      4)The key point to do Time Warp is to find the Global Virtual Time(GVT) which is the minimum.

    2.Difference between Fault Tolerance and Load Balancing.

      1)Falult Tolerance means to solve the condition that some machine is crushed, the models on it should be changed to another machine.

      2)When the condition is detected, the simulation is stopped first, then the models will be re-mapped.

      3)Load Balancing is a dynamic procedure, the utilization is the key parameter to decide whether to do the Load Balancing.

      4)Both Falut Tolerance and Load Balancing need to re-map the models.

      5)The aim of Load Balancing is to optimize the simulation, and the cost of Load Balancing mainly comes from Delay and Less Efficiency of network.

    3.How can Ujean and Amr get the value of time out.

      1)The model of network is constructed to do the experiment.

      2)The calibration is done by the experiment. Then from the curve of Performance and Runtime(y axis) with Time out(x axis), the most suitable time out value will be found.

      3)Before the simulation, we should do the Calibration; Validation; Performance Analysis to make sure about the system parameters of simulation model is nearly to the real ones, ie: time out.

    4.Flatten, Distributed, Activity Trace, Time Warp of DEVS modeling.

      1)The devs model could be constructed in the first two styles, Different styles are needed according to different models.

      2)Flatten is an important direction to transform DEVS models from hierachy to flattened, and then the simulator of atomic model could be simple and powerful.

      3)Activity Trace is useful in the simulation like forest fire, the orginal model of it could be traced out.

    5.The main task next step.

      1)The Partitioning is key problem in distributed DEVS, then there are some solutions in application. And we can abstract it into the model level, to study the partitioning from the Domain Spefication Model. Then the more useful and general model could be constructed to do partitioning.

      2)Flatten DEVS model seems to make good sense to the modeling.

      3)To learn the detail of the model transformation, try to transform the model from DEVS to BOM each other.

      4)To learn the detail of the implementation of DEVS/HLA.



One-to-One Meeting on Friday Morning from 10:00am to 11:00am (23 January 2009)

    Talked about Devs Flattening:

    1.Do the Direct Connecting first.

      1)Find the most simple coupled DEVS model which doesn't contain the coupled DEVS model.

      2)Take all the atomic DEVS model to the upper level coupled DEVS model.

      3)Reconnecting the models in the upper level coupled DEVS model.

      4)Do the process like this recursively.

    2.Flattening DEVS.

      1)After the Direct Connecting, the only one coupled DEVS which contain lots of atomic DEVS models is left.

      2)Generate the new of State Class for the flattened Coupled DEVS model .

      3)Add the Status Variable X and Y to all the atomic DEVS models.

      4)Generate the Atomic DEVS model to replace the Coupled DEVS model.

      5)Add the Functions MapY2X which do the data transferring instead of ports in atomic DEVS model befroe to the coupled DEVS model.

      6)Rewrite the internal transition function external transition function and the time advance function based on the Closure of coupled DEVS model.

    3.The Z and select functions.

      1)Z and select functions are not considered in the Bill's implementation.

      2)They should be implemented in the flatten DEVS model.

      3)Z function and select functions are defined by users, they should be melted into the expression into the flatten DEVS model.



One-to-One Meeting on Friday Morning from 10:00am to 11:00am (13 February 2009)

    Talked about the Testing Models for DEVS Flattening:

    1.Use the Circuit Model to be the High Interaction Models to test DEVS Flattening.

      1)Build the Semi-Summator based on the XOR and AND gates.

      2)The experiment show that the Direct Connection improve the simulation a little.

      3)The experiment show that the Flattening improve the simulation a lot.

    2.Build more complex testing model like DATAPATH.

      1)Build the 8 bit Adder, using the RIPLE CARRY, CARRY LOOKAHEAD.

      2)Build the 8 bit Multiplier.

      3)Build the 8 bit OR, XOR, AND, NOT.

      4)Build the SPLITTER and JOIN model.

      5)Build the memory to be the "REGISTER", using the model of "FLIPFLOP".

      6)Build the ALU at last and maybe the DATAPATH.

      7)Use the reference of
    3.About the paper.

      1)Read the publications before, especially the paper "Distributed Simulation of Hierarchical DEVS Models: Hierarchical Scheduling Locally and Time Wrap Globally".

      2)Focus on the flattening itself, find the research had been done and cover all the research around the DEVS flattening which had been published.
Maintained by Bin. Last Modified: 2009/02/13 22:19:14.