Improvement of Distributed Virtual Environment (DVE) performance

Document Sample
Improvement of Distributed Virtual Environment (DVE) performance Powered By Docstoc
					                                                                  (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                            Vol. 9, No. 3, 2011

         Improvement of Distributed Virtual Environment
                     (DVE) performance
                            Olfat I. EL-Mahi                                                           Hanan Ali
                   Computer Graphics department                                             Computer Graphics department
                      IRI institute- MuCSAT                                                    IRI institute- MuCSAT
                      Borg EL-Arabe, Egypt                                                      Borg EL-Arab, Egypt
                     Olfat.ibrahim@gmail.com

                           Walaa M. Sheta                                                             Salwa Nassar
                  Computer Graphics department                                                 Electronic Research Institute
                     IRI institute- MuCSAT                                                             Cairo, Egypt
                      Borg EL-Arab, Egypt                                                          snassar@narss.sci.eg
                     wsheta@mucsat.sci.eg

Abstract— Distributed virtual Environment enables multiple                   Because different avatars participating in one DVE System.
users to interact with each other over a network. Due to the                 DVE Systems performance can face some difficulties:
availability of high bandwidth and fast graphics cards, these
systems grow in term of number of users, Scene Complexity and                    1.   Different usurer’s background and experiences:
interactivity. However, the issue of how efficient the system scales                  different applications can deal with different user aims
as the number of users increase is major problem that DVE faced                       and responses and background experiences. This
since their inception.                                                                should effect on user communication rate with other
In this paper, we propose a new method in order to improve DVE                        users. Also the application itself may have a lot of
performance using Data-Mining. A widely used mining technique                         different options to switch between. This could cause
is markov chain model, which depend on predicting the future                          the system to be heavy or even stop responding due to
user moves based on the history of his previous visits to the DVE                     the multiple switches.
System. This will help decreasing information exchange between                   2.   Different network: users can interact through different
users, which should in turn enable improvement in the walk                            networks technologies with various speeds and
through in the distributed virtual environment system (DVE).                          capabilities. This variation could have some effect on
Keywords-component; HLA, DVE, Markov chan model.                                      the DVE system performance.
                                                                                 3.   Different Computer resources: the performance of
                          I.     INTRODUCTION                                         computer device can change dramatically depending
                                                                                      on its memory size and speed, its visual graphic card
   Due to The Major production of high performance graphic                            type and capability and its processor.
cards that is offer a good frame rate rendering and availability
of cheap networks with high bandwidth the field of Distributed
Virtual Environment (DVE) system has attracted a greater                          According to this variation on user’s platforms, producing
interest from researchers in the past few years. This was in                 DVE System with high performance and wide scalability
order to fill in the argent need for customers demands raised in             become a great challenge. One of the Key issues to achieve this
this field. These systems permit more than one user users,                   challenge is by targeting Network Traffic Reduction methods.
working on different Locations, which are interconnected                     Reducing the number of messages needed to be transferred
through different networks to interact in a shared virtual world             between users will directly scale the number of users that can
[1]. Its aim is to simultaneously allow the participating users to           participate and receive accepted Quality of Service (QOS). A
share one virtual world, interact with it and with each other and            number of solutions were exhibit for this problem like dead-
give them the feeling of a real experience. It does that by                  reckoning technique which offers some level of independence
rendering images of the virtual environment and the updating                 to users [2, 3]; broad cast or multi cast solutions which allow
information of other users participating on it as long as the user           some way of decreasing number of exchanged messages and
continues into his navigation experience. Each user                          keep the system consistency [4-6]. Another method focused on
participating in the system is represented by avatar. Because                load balance between servers to maintain high system
system provide real time visual interacting, Each avatar on the              performance [1].
system will not only be responsible on performing the
computes for his own behavior and publish them to the                             This paper demonstrating a new methodology based on
network but it must accurately represent all other entities                  recognizing navigation history of different users types and
participating in the DVE.                                                    background to perform some common behaviors or targets
                                                                             while walking in a certain virtual world in different time
                                                                             intervals during the day. Building some information bank for
      Identify applicable sponsor/s here. (sponsors)



                                                                       287                             http://sites.google.com/site/ijcsis/
                                                                                                       ISSN 1947-5500
                                                                     (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                               Vol. 9, No. 3, 2011
different users should later help in predicting there next step.             this technique the system still acquired to provide information
This should help in decreasing the need for data exchange                    about current and future navigators positions. Increasing in the
through the Network. And as a result will help improving                     number of users or processes in the DVE system lead to
System QOS.                                                                  exhaustion of available network bandwidth due to the
                                                                             quadratic increase in network traffic. We try to identify a
   A prototype built using Virtual City Simulation to apply                  method to recognize a pattern (e.g. a sequences), learned from
multi-servers architecture using data distribution management                users database using Markov Chain model. Predicting what
(DDM) service provided onto the High Level Architecture                      would be the frequent directions for each user. That should
(HLA) protocol, which our system built, based on it. But even
                                                                             save a lot of bandwidth because the user will not have to send
with this technique the system still acquired to provide                     his current position update or spread other users location
information about current and future navigators positions.                   unless it is came with different frequencies or directions then
Increasing in the number of users or processes in the DVE                    his usual.
system lead to exhaustion of available network bandwidth due
to the quadratic increase in network traffic. We try to identify a                            III. MARKOV CHAIN MODELS
method to recognize a pattern (e.g. a sequences), learned from
users database using Markov Chain model. Predicting what                        Markov chain concerns about a sequence of random
would be the frequent directions for each user. That should                  variables. These variables are corresponding to the states of a
save a lot of bandwidth because the user will not have to send               certain system, in such a way that the state at one time epoch
his current position update or spread other users location unless            depends only on the one in the previous time epoch. It could
it is came with different frequencies or directions then his                 be defined in the following matter:
usual.                                                                       Given some sequence x of length L, we can ask how probable
   Markov chains have recently been used to model user                       the sequence is given our model. For any probabilistic model
navigational behavior on the World Wide Web (WWW) [7, 8].                    of sequences, we can write this probability as:
In our work, a framework will be implemented for building
distributed simulations. This framework based on High level
architecture (HLA) protocol for applying client-server                                                                                         (1)
architecture and then a method will be proposed for
constructing a Markov model of a users paths prediction based                         Key property of a (1st order) Markov chain: the
on past visitor behavior. Using it should assist users to navigate           probability of each Xi depends only on Xi-1:
the DVE system.
                      II.   RELATED WORK
   Earlier systems, such as NPSNET [4, 9] and DIVE [10], are                                                                                   (2)
implemented in a peer-to-peer architecture. This approach has
minimal communication overheads, but may not scale well to
handle many simultaneous clients due to the saturation of                        For example consider a user navigating into a city
network bandwidth in handling broadcast or multicast                         consists of 17 streets. Each street labeled with alphabetic
messages from the clients. To improve scalability, systems                   character to trace user navigation into it. The markov chain
such as Brick Net [11], Community Place [12]and MASSIVE-                     model will be defined in the following manner:
3[13], are implemented in client-server architecture. With this                   •    A set of states: representing by the streets labels “A,
approach, each client sends messages to the server for further                         B, …, W”
propagation to other clients and/or server in the same VE. The                    •    A set of transitions with associated probabilities:
advantages of this approach are that the server may perform                            represent the user move from one street to the next
message filtering to minimize the amount of messages needed                            one.
to be handled by each client and to be propagated through the
network. The major limitation, however, is that as the number                Consider the user move in the path “Fig. 1”:
of clients accessing the VE increases, the amount of messages                         The probability for this path will be calculated with
needed to be sent among them increases dramatically. The                     the following equation:
server loading in managing the VE and handling the messages
also increases significantly. Another problem is that the server
may potentially become a single point of failure.
                                                                                                                                               (3)
A distributed VE system with a multi-server architecture could
solve these problems. The VE may be partitioned into regions,                         The Markov prosperity specifies the probability of a
and each of which is assigned to a separate server, distributing             state depends only on the previous state. But we can build
the workload among them. This may also prevent the single                    more “memory” into our states by using higher order Markov
point of failure problem if clients can be dynamically                       chain. In a k-th ordered Markov model.
connected to different servers. Systems adopting this approach
include RING [14], Net Effect [15] and Citation [16].
                                                                                                                                               (4)
   In our proposed DVE System we apply multi-servers
architecture using data distribution management (DDM)
service provided onto the High Level Architecture (HLA)
protocol, which our system built, based on it. But even with
                                                                      288                               http://sites.google.com/site/ijcsis/
                                                                                                        ISSN 1947-5500
                                                                            (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                      Vol. 9, No. 3, 2011
                                                                                                          Table 1. Different Building types.




        Figure 1. Example of user movement steps in the path


i.e., ith value is independent of all but the previous k value.
         For the same user path in the previous example the
3rd order markov chain the path probability will be:                                A. The Shared Virtual Environment
     Pr(U | BSCSO) = Pr(U | CSO )                                                   Creating a virtual environment can take as long as creating the
                              Where i = 6, K = 3                      (5)           whole DVE system. It is considered to be the core of any DVE
                                                                                    System. Also Buying Virtual Environments to be used as a
                 IV.    Our proposed DVE system                                     testing data set costing thousands of dollars. So we proposed
                                                                                    virtual environment system constructor (Vcit) to create the
The proposed DVE system mainly consists of three parts:                             simulated virtual Environment. It is creating virtual city
Virtual System, Distributed Virtual Environment and Data-                           According to the standard city design rules[17, 18].
Mining Engine. They incorporate together ‘like in Fig (2)’ to
simulate an experience of real time interaction between
participate users in the chosen shared virtual environment. And                     It automatically generates cities include the standard main
try to provide the best quality of service available for each user.                 components” streets, highways, intersections, blocks, areas
                                                                                    “fig. 3”. The cities generated with different Size, Distribution,
                                                                                    and layout and it deal with twenty different building types
                                                                                    “Table. 1”.




                                                                                                        Figure 4. The two main parts of the Vcit



                                                                                    The Vcit generator consists of two main parts “Fig. 4”:

       Figure 2. The general structure of proposed DVE system                       1) Data Set Generator: Determine City Size, Distribution,
                                                                                       layout and properties for the city and each building on it.
                                                                                       It is generated using Sybase DBMS. The output is
    Each of the previous parts will be discussed briefly next,                         generated as a text file comes in two different formats
giving a general idea of their mechanism and how they work.                            depending on the chosen layout for the virtual city “Fig.
                                                                                       5”:
                                                                                       o Using different texture.
                                                                                       o Using different RGB




         Figure 3. The standard main components of any created city



                                                                             289                                   http://sites.google.com/site/ijcsis/
                                                                                          Figure 5. The different possible layout for virtual city using Vcit
                                                                                                                   ISSN 1947-5500
                                                                 (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                           Vol. 9, No. 3, 2011
                                                                              •    Collision Detection: Detect collisions between users
                                                                                   and each other’s and between users and the virtual
                                                                                   city.

                                                                              •    User Registration: record the registration for each
                                                                                   user log onto the system.

                                                                         C. Data-Mining Engine




            Figure 6. The scene graph structure for the Vcit

2) Layout builder: Produce the Virtual City Lay out
   Depending on the Generated text file. The virtual city
   represented in a scene graph structure “Fig. 6” It is
   created using OpenGL Performer.




                                                                                          Figure 8. The data-mining engine architecture

                                                                             The aim of adding data-mining engine to the DVE system
                                                                         was to learn from users’ data and get information used to
                                                                         improve the system performance. “Fig. 8” shows our data
                                                                         mining engine architecture. It is working in the following
                                                                         manner:

                                                                              •    Request Handler: responsible on handling updates
                                                                                   sent and received by users then send it to the path
                                                                                   recorder.
                                                                              •    Path recorder: is responsible on recording each user
                                                                                   path and send it to the log data manager.
           Figure 7. The proposed Graphic station Architecture
                                                                              •    Log data manger: responsible on recording users’
B. Distrubited Virtual Environment                                                 paths in a suitable format to be handled by the mining
                                                                                   unit. Then send it to the mining unit.
   In the construction of the Distributed Virtual Environment
we depended on the High Level Architecture (HLA) protocol                     •    Mining unit: mine the recorded paths and get the
to be used as communication protocol allow different users                         frequent sequences for each user type and send it
share the Virtual Environment.                                                     back to the store manager.
                                                                              •    Storage manger: record the frequent sequences for
   The proposed Graphic station Architecture “Fig. 7” offer                        each user type and return it back to the federate
10 services six of them implemented using the HLA Protocol                         responsible for the server and it is in turn send it to
[19] these services are: Federation Management, Declaration                        all users to use it in next step prediction.
Management, Object Management, Ownership Management,
Data Distribution Management and Time Management.
                                                                                     V.      The frequent way mining procedure
The other four services are built using the OpenGL Performer:
                                                                         This procedure work into three phases
    •   Scene Manager: Updates from Object Management.                        A. Data collection phase.
                                                                              B. Building Local Profile.
    •   Object Management: receive position and status and
        send it to scene manager.                                             C. Predict next step.

                                                                  290                                  http://sites.google.com/site/ijcsis/
                                                                                                       ISSN 1947-5500
                                                                   (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                             Vol. 9, No. 3, 2011
    These phases aim to detect the frequent way sequences for              3. The proposed paths
different users’ types then use these patterns it predicting
future movement steps. This should help decreases the data                          Each client or user suppose to start from home
exchange between users.                                                    location which is a random location at one of the following
                                                                           building types: Residential, Residential 2, Residential 3,
A. Data collection phase:                                                  Residential 4, Villa 1, Villa 2 “Fig. 10”. User walks into the
                                                                           street until he reach an intersection then three turns are
   The data collection phase based on the distribution for the             available and picked up randomly: Left turn, Right turn and
used city, the chosen aviators’ kind and the proposed paths.               Straight throw. User stops when he reaches one of his target
                                                                           locations according to his type.
1. City Distribution




                                                                                             Figure. 10 the proposed user routes
         Figure. 9 The city Distribution used in our experiments
                                                                           B. Building Local Profile
          The virtual city used in our experiments consists of:
17 Streets, 6 intersections and four highways surrounding it.                   In order to capture user-moving pattern, a movement log
Each Street labeled with an Alphabetic character to be able to             is needed. A movement log contains pair of (old location, new
trace user path “Fig. 9”. 170 building from 20 different                   location) .in the beginning of a new path the old location is
building types was produced to occupied Main streets based                 null for each user, a moving sequence {(O1, N1), (O2, N2)…
on real city.                                                              (On, Nn)} Can be obtained from movement log. Knowing that
                                                                           on = Nn-1.
2. Avatars kind
                                                                                Consider the example “Fig. 11” where number next to
         Eight different avatars type was proposed based on:               each link represents the sequence of moving for a user. Thus,
Age, gender, Work or education background and Time                         the moving log contains the moving path for that user. The set
interval of the day. And based on each user type the targets               of maximal moving sequence for this moving path output by
buildings specified, Table (2).                                            the using MM algorithm will be {ARBSOUOSBR}.

                                                                                Once the movement log is generated, we shall convert the
                  Table: (2) the Eight different avatars type
                                                                           log data into multiple subsequences, each of which represents
        User Type                      Proposed Target Locations           a maximal moving sequence. After maximal moving
        Homemakers                     markets or Clubs                    sequences are obtained, we then map the problem of finding
        Students                       Schools or Clubs                    frequent moving patterns into the one of finding frequent
        Tourists, Researchers          Museums, Clubs or Hotels
        Government employees           Government Buildings
                                                                           occurring consecutive subsequences among maximal moving
        Doctors or Nurse               Hospitals                           sequences. A sequence of k movements is called a large k-
        Accountants                    Banks                               moving sequence if there are a sufficient number of maximal
        Police officers                Police Offices                      moving sequences containing this k moving sequence. Such a
        Postal workers                 Post Offices                        threshold number is called a support. Note that after large
                                                                           moving sequence that is not contained in any other moving
                                                                           patterns. For example, suppose that (AB, BC, AE, CG, GH) is
                                                                           the set of large 2-moving sequences and (ABC, CGH) is the
                                                                           set of large 3-moving sequences. Then, the resulting user
                                                                           moving patterns are (AE, ABC, and CGH). User moving
                                                                    291                                http://sites.google.com/site/ijcsis/
                                                                                                       ISSN 1947-5500
                                                                   (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                             Vol. 9, No. 3, 2011
                                                                           the moving pair (Oi, Ni) is one of the target locations for the
                                                                           User Kind Running. Otherwise, Ni is appended into Y (in line
                                                                           13 of algorithm MM) and the occurrence count of (Oi, Ni) is
                                                                           updated online in the database (in Line 15 of the algorithm
                                                                           MM).




            Figure 11. example of user moving sequance

patterns are associated with the areas that user frequently
travel in a DVE system. The overall procedure for mining
moving patterns is outlined as follows:

1.   Finding Maximal Moving Sequences

     As pointed out earlier, a moving pair, (old StreetNo, new
StreetNo), is generated in a movement log for each                                    Figure. 12 the maximal moving sequence algorithm
registration procedure. Given a moving sequence {(O1, N1),
(O2, N2)… (On, Nn)} Of user, a map for multiple                            An example execution scenario by algorithm MM. for the
subsequences will be produced, each of which represents a                  input in “Fig. 13” is given in Table 3. Knowing that N is the
maximal moving sequence.                                                   Start location, E is one of the target locations, and this Path
                                                                           sequence is applicable for user from type 1 or 6.
    First, a moving sequence {( O1,N1}),( O2,N2),…(                               Table. 3 An example execution scenario by algorithm MM
On,Nn)} can be obtained for each user from the movement
log, where pairs of (Oi,Ni) are sorted by time. Then ,
algorithm MM (standing for maximal moving sequence),
whose algorithmic form is given below “Fig. 12” , is applied
to moving sequences of each user to determine the maximal
moving sequence of that user and update the occurrence count
of moving pairs during registration procedure.

     In algorithm MM, Y is used to keep the current maximal
moving sequence and F is a flag to indicate if a node is
revisited. Let DF denote the Database to store all the resulting
maximal moving sequence.

     In addition, S is the home location of the user. According
to the roundtrip considered, the selection of S is a geography
area, which has been chosen randomly assuming it is its home
building of the user. In order to capture the complete traveling
sequence, algorithm MM outputs a maximal moving sequence
to DF until a target location reached. A target location has
been adapted for each user due to its assumed needs according
to the figure.

     In line one of algorithm MM, a Set User Type, and targets
according to it are initialized. Then in line two, some
parameters are initialized. Then, moving sequence is scanned
in line 3. A maximal moving sequence will be explored (from                               Figure. 13 An example execution scenario by
Line 15 to line 19 of algorithm MM) if MM finds that Ni in
                                                                    292                                http://sites.google.com/site/ijcsis/
                                                                                                       ISSN 1947-5500
                                                                    (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                              Vol. 9, No. 3, 2011
2.   Finding Large moving Sequences

         With the maximal moving sequences obtained, we                       Algorithm LM/* Algorithm for finding large moving sequences */
next determine the large moving sequences. A large moving                     Input: A set of w maximal moving sequences of a DVE user of a certain
sequence can be determined from all maximal moving                            type user.
sequences of each individual user type based on its                           Output: Large moving sequences of the DVE user.
                                                                              Begin
occurrences in those maximal moving sequences. Use intra-                     1. Determining L2 = { large 2- moving sequence } from moving pairs in C2;
sequence count to mean the number of occurrences of a                         2.    for (K= 3 ; Lk-1 != 0 ; k++)
moving sequence within a maximal moving sequence and                          3.     begin
inter-sequence set of a moving sequence to mean the set of                    4.            Ck = Lk-1; /* Generating Ck from Lk-1*Lk-1*/
maximal moving sequences, which contain that moving                           5.             for w maximal moving sequence S
sequence.                                                                     6.              begin /* Calculating the intra-sequence count of Ck within
                                                                              S*/
                                                                              7.                 intra-sequence = subsequence ( Ck , S );
         The count of a large moving sequence is the sum of                   8.                  if (intra-sequence >0)
intra-sequence counts from its inter-sequence set. For the                    9.                       Including S into intra-sequence set ;
                                                                                                   /* sum of occurrence counts in a intra-sequence set */
example in Table 4, the intra-sequence count of GB in                         10.                      for all candidate c E intra-sequence
{ABCGBCGBA} is two and that in {ABGBA} is one.                                11.                       c.count = c.count + c. intra-sequence;
Moreover, the inter-sequence set of GB is {{ABCGBCGBA},                       12.           end
{ABGBA}}. Hence, the count of GB is the sum of intra-                         13.          Lk = { c E Ck/ c.count >_ support } ;
sequence counts in its intra-sequence set (i.e., 2 (i.e., intra-              14.     end
sequence count in ABCGBCGBA) + 1 (i.e., intra-sequence                        end
count in ABGBA) = 3).
          Table. 4 The count of a large moving sequence                                         Figure. 14 the large moving sequence
                                                                            C.        Predict next step.

                                                                                      We can see that Before Using DM approach User had
                                                                            to send update with his location at every step. Now if an
                                                                            existing pattern User only needs to send his start point and
                                                                            pattern number. If not existing pattern and going in straight
                                                                            lines: user sends his start location and the point at the start of
                                                                            each new street along with the street number.


          To cope with this problem, we develop algorithm LM                                      VI.     Experiments Setup:
(standing for large moving sequence) for the determination of
large moving sequences ‘Fig 14”. Let Lk represent the set of                   A System Architecture based on networked servers was
all large K-moving sequences and Ck be a set of candidate K-                then developed as stander for DVE systems. Using HLA as its
moving sequences.                                                           main standard, a developed simulation tool written in C++,
                                                                            which model the behavior of avatars in the generic DVE
          As will be show, the initial candidate set generation,            system.
especially for L2, is the key issue to improve the performance
of data mining. Since occurrence counts of moving pairs, i.e.,                 The DVE system is composed of S interconnected servers
C2, were updated online in the data Collection phase, L2 can                and n avatars under the main HLA Server (called agent name
be determined by proper trimming on C2 efficiently (line 1 of               service or Federation Manager) and it manages the whole
algorithm LM), showing the advantage of having online                       system. A multi-threaded server implemented each on a
update in algorithm MM. Also, note that Ck can be simply                    different, dedicated computer. Each of this threads
generated from Lk-1*Lk-1 (line 4 of algorithm LM). For                      communicating with clients’ computers. Each client controls
example, with the set of L2 Being {AB, BK}, a C3 will be                    the behavior of a single avatar and it is implemented as a
{ABK}. As explained above, the occurrence count of each k-                  multithreaded application, which uses to manage the user
moving sequence is the sum of intra-sequence count (from line               information (current position, latency, etc).
5 to line 9 in the algorithm LM) in its inter-sequence set (i.e.,
line 10 and line 11 in the algorithm LM). The occurrences of                   The main HLA server was installed on DELL server with
each k-moving sequence in Ck are determined for the                         Intel Xeon processor, CPU 3.20 GH with 4.00 GB Rams. Each
identification of Lk. After the summation of the occurrence                 Server was represented by IBM Computer with Intel Pentium
counts in the inter-sequence set from line 10 to line 11 in the             4 processor, CPU 2.4GHZ with 2.00 GB Ram. The application
LM algorithm, those k-moving sequences with counts                          was installed on three Sun servers with AMD OPTERON
exceeding the support are qualified as Lk ( line 13 of                      processor 248, CPU 2.19 GHZ with 2.00 RAM.
algorithm LM). Notice that those large k-moving sequences
are obtained from ŵ maximal moving sequences of that user,
showing the incremental mining capability of algorithm LM.


                                                                     293                                 http://sites.google.com/site/ijcsis/
                                                                                                         ISSN 1947-5500
                                                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                    Vol. 9, No. 3, 2011
                                                                                    applying the data mining technique is much more then when
                                                                                    not applying it. In addition, the total number of messages sent
                                                                                    by all users with applying data mining technique is much less
                                                                                    then without applying it.

                                                                                    C. Experiment 3:
                                                                                             The main aim of this Experiment is to test the
                                                                                    efficiency and requirement of the proposed approach to
                                                                                    improve the DVE system by testing the performance of the
                                                                                    Real-time rendering for the Hardware Graphics Pipeline3D.
  Figure 15: shows that No. of updated steps sends via network increase
                        with the number of clients
                  VII. Experiments and results:

A. Experiment 1:
    In this experiment, we study the effect of number of
aviators on the performance of DVE. As shown in table 5, the
number of clients ranges from 50 to 500. The number of
servers is fixed at two. The client distribution pattern is fixed
at uniform. Figure 15 shows the no of packets needed in
                                                                                                          Figure 20. rendering process
clients update during the whole run versus the number of
clients at each. We can see that no of updates packets sends                        If an application is displaying 15 fps, it is considered real-
via network increase with the number of clients.                                    time. The architecture of the real-time rendering pipeline can
                                                                                    be divided into three conceptual stages “Fig. 21”
               Table 5. Experiment 1 setup
           Parameter               Description
           No. of Servers          2
           No. of Users            50-500
           No. of user types used 8

B. Experiment 2:




                                                                                          Figure 21. the real-time rendering pipeline conceptual stages

                                                                                    The application stage: is driven by the application where “it
                                                                                    begins the image generation process that results in the final
                                                                                    scene of frame of animation? The application is implemented
     Figure 16: shown the total no of steps generated with applying and             in the software thus giving the developers total control over
             without applying DM approach is almost the same.                       the implementation in order to change the performance.
          The purpose of this experiment was to study the
scalability of frequent walking sequence methodology. By                            The geometry stage: is responsible for the majority of the
applying, the most frequent sequence results provided from                          per-polygon operations or per-vertex operation; it means that
our mining technique to be tested to determine if it could                          this stage computes what is to be drawn, how it should be
improve the quality of the user interaction with the virtual                        drawn, and where it should be drawn.
reality system.
                                                                                    The Rasterizer Stage: Once all of the necessary steps are
         Figure 16 shows the no of packets sent by clients in                       completed from the two previous stages, all the elements,
updating during the whole run versus the number of clients at                       including the lines that have been drawn, models that have
each run specified for each user type. As showing the results                       been transformed into what are that were done from those
given with applying of the data mining technique is, better                         stages are ready to enter the rasterizer stages. Rasterizer stage
than the one without applying it. In addition, the total no of                      means turning all of those elements into pixels, or picture
steps performed by both users is almost the same.                                   elements, and adding color onto them.

The Figures 17 18, 19 ,tables 6,7 shows that the total number                       Each of the previous states accomplished in a certain place at
of packets needed not to be sent via the network with                               the computer with a certain arrangement which called




                                                                              294                                  http://sites.google.com/site/ijcsis/
                                                                                                                   ISSN 1947-5500
                                                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                    Vol. 9, No. 3, 2011
Texture Mapping and Z-Buffer. Application and geometry                                       Transactions on Parallel and Distributed Systems,
stages accomplish in the Central Processing Aria (CPU) of                                    2005: p. 637-649.
the computer. But the rasterization stage accomplished in the                        2.      Frécon, E. and M. Stenius, DIVE: A scaleable
Graphical Processor Unit (GPU).                                                              network architecture for distributed virtual
                                                                                             environments. Distributed Systems Engineering,
In this experiment we will change the window size for the                                    1998. 5: p. 91.
user which in turn will change the no of pixels to be mapped                         3.      Harvey, E., R. Schaffer, and S. McGarry. High
then will measure the changing of no of fps due to the                                       performance fixed-wing aircraft simulation using
increase of the no of users with a fixed graphical card type                                 SIMNET protocols. 1991.
Nvidia Quadro 4400.                                                                  4.      Falby, J., et al., NPSNET: Hierarchical data
                                                                                             structures for real-time three-dimensional visual
The result “Fig 22” shows that without using frequent way                                    simulation. Computers & Graphics, 1993. 17(1): p.
mining approach, with two different windows size, a boatel                                   65-69.
nick happened decreasing the no of fps when no. of users                             5.      Lee, D., et al., ATLAS: a scalable network
reach 7 users. But when using frequent way mining approach                                   framework for distributed virtual environments.
number of users rises to 50 users. Data transferred between                                  Presence: Teleoperators and Virtual Environments,
users and affect it. Which show the necessity for try to                                     2007. 16(2): p. 125-156.
decrease the dependent on the exchanged messages between                             6.      Zou, L., M. Ammar, and C. Diot. An evaluation of
users to update the scene, which should in turn decrease the                                 grouping techniques for state dissemination in
chance for any system boatel nicks occurrence.                                               networked multi-user games. 2002. IEEE.
                                                                                     7.      Sarukkai, R., Link prediction and path analysis
                                                                                             using Markov chains1. Computer Networks, 2000.
                                                                                             33(1-6): p. 377-386.
                                                                                     8.      Eirinaki, M., M. Vazirgiannis, and D. Kapogiannis.
                                                                                             Web path recommendations based on page ranking
                                                                                             and Markov models. 2005. ACM.
                                                                                     9.      Zyda, M., et al., The software required for the
                                                                                             computer generation of virtual environments.
                                                                                             Presence, 1993. 2(2): p. 130-140.
                                                                                     10.     Carlsson, C. and O. Hagsand. DIVE A multi-user
                                                                                             virtual reality system. 2002. IEEE.
                                                                                     11.     Singh, G., et al. BrickNet: sharing object behaviors
                                                                                             on the Net. 1995. Published by the IEEE Computer
Figure 22. the number of frames per second changing with the number of users                 Society.
                                                                                     12.     Lea, R., et al. Community Place: architecture and
                VIII. Conclusion and Future work
                                                                                             performance. 1997. ACM.
                                                                                     13.     Greenhalgh, C., J. Purbrick, and D. Snowdon. Inside
         Navigation in large-scale VEs may be Complex                                        MASSIVE-3: flexible support for data consistency
task. Many computer resources mostly depending on                                            and world structuring. 2000. ACM.
information send exhaust between users. By applying DM                               14.     Funkhouser, T. RING: a client-server system for
technique improve to the DVE system produced by reducing
                                                                                             multi-user virtual environments. 1995. ACM.
needed information exchange without effecting the users’
information. Which in turn improve the service provided                              15.     Das, T., et al. NetEffect: a network architecture for
between users cause of reduce the resources needed in                                        large-scale multi-user virtual worlds. 1997. ACM.
information exchange.                                                                16.     Hori, M., et al. Scalability issues of dynamic space
                                                                                             management for multiple-server networked virtual
                                                                                             environments. 2002. IEEE.
        Our future work is plan to enhance the frequent
                                                                                     17.     Darken, R. and J. Sibert. Wayfinding strategies and
walking sequence interface with interactive visual previews
of recommended routes and developing a new web usage                                         behaviors in large virtual worlds. 1996. ACM.
mining technique for deriving frequent path traversal patterns                       18.     Bowman, D., et al., 3D user interfaces: theory and
in web environments.                                                                         practice. Vol. 2. 2005: Addison-Wesley Boston
                                                                                             (MA).
                                                                                     19.     Kuhl, F., R. Weatherly, and J. Dahmann, Creating
                             REFERENCES                                                      computer simulation systems: an introduction to the
                                                                                             high level architecture1999: Prentice Hall PTR
1.        Morillo, P., Improving the performance of                                          Upper Saddle River, NJ, USA.
          distributed virtual environment systems. IEEE




                                                                               295                             http://sites.google.com/site/ijcsis/
                                                                                                               ISSN 1947-5500

				
DOCUMENT INFO
Shared By:
Stats:
views:88
posted:4/11/2011
language:English
pages:9
Description: Distributed virtual Environment enables multiple users to interact with each other over a network. Due to the availability of high bandwidth and fast graphics cards, these systems grow in term of number of users, Scene Complexity and interactivity. However, the issue of how efficient the system scales as the number of users increase is major problem that DVE faced since their inception. In this paper, we propose a new method in order to improve DVE performance using Data-Mining. A widely used mining technique is markov chain model, which depend on predicting the future user moves based on the history of his previous visits to the DVE System. This will help decreasing information exchange between users, which should in turn enable improvement in the walk through in the distributed virtual environment system (DVE).