Simulation of DoD Collaboration Services

  • Upload
    agriggs

  • View
    213

  • Download
    0

Embed Size (px)

Citation preview

  • 8/8/2019 Simulation of DoD Collaboration Services

    1/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Simulation of DoD Collaboration Services

    Aaron Griggs

    11 December 2007

    SYS 611 Modeling and Simulation

    Stevens Institute of Technology

    1

  • 8/8/2019 Simulation of DoD Collaboration Services

    2/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Table of Contents

    Background ................................................................................................................................. 3

    Model .......................................................................................................................................... 3Simulation ................................................................................................................................... 3

    Text Chat ................................................................................................................................. 4Attributes............................................................................................................................. 4Users ................................................................................................................................... 4

    Simulated Time ................................................................................................................... 4

    Results ................................................................................................................................. 4

    Audio Collaboration ................................................................................................................ 7Attributes............................................................................................................................. 7

    Users ................................................................................................................................... 7

    Simulated Time ................................................................................................................... 7Results ................................................................................................................................. 7

    Video Collaboration .............................................................................................................. 10

    Attributes........................................................................................................................... 11Users ................................................................................................................................. 11

    Simulated Time ................................................................................................................. 11

    Results ............................................................................................................................... 11Aggregate .............................................................................................................................. 14

    Attributes........................................................................................................................... 14

    Users ................................................................................................................................. 14

    Simulated Time ................................................................................................................. 14Results ............................................................................................................................... 14

    Conclusion ................................................................................................................................ 18

    2

  • 8/8/2019 Simulation of DoD Collaboration Services

    3/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Background

    The Department of Defense like other large enterprises is adopting collaboration services to

    promote better communication and take advantage of a distributed workforce. The key

    collaboration services include: text chat, audio, video, white boarding, file sharing and webconferencing (multi-user chat).

    The first deployment of collaboration services will be centrally hosted at the Defense Enterprise

    Computing Center (DECC) in Columbus, Ohio consisting of three services: text chat, audio and

    video collaboration. The collaboration services (i.e., software and hardware running at theDECC) are being outsourced and can be scaled up as necessary. However, the network

    performance is the DoDs responsibility and requires significant investments if the existing

    infrastructure cannot support the collaboration load.

    The expected concurrent user load consists of the following:

    5000 users of text chat 500 users of audio 50 users of video

    Prior to contracting for the collaboration services, the DoD is interested in understanding the

    ability of the network to support delivering the services to the end users. Two network

    performance metrics are of interest: response time and utilization. A response time of 0.1seconds or less and network utilization of 80% and below is considered acceptable.

    Prior to investments in the services, the DoD needs to understand if the existing network is able

    to support the services. The assumption is the DECC is connected to regional Points of Presence

    (POPs) via an OC3, 155 megabits per second (Mbps). The end users (consumers) are connectedto regional POPs via a DS3, 45 Mbps.

    The DoD has chosen a modeling and simulation approach to see how the network is affected by

    the three collaboration services.

    Model

    A network model consisting of the centrally located services was simulated in OPNET. Theconsumers are assumed to be located at STRATCOM in Omaha, Nebraska, connected through a

    regional POP located in Kansas City, Missouri.

    A single consumer location was chosen for simplicity. In reality, the users would be distributed

    all over the US (assuming continental US usage at this time). However for aggregation ofnetwork links, a single location suffices.1

    SimulationEach service was simulated with a DS3 link from KC to STRATCOM as well as replacing that

    link with one well under capacity for the service load to see the impact to service delivery. The

    network model is shown in the diagram below.

    1 Note the assumption is multicast is not used. Future models and simulations should test the affects of leveraging

    multicast for UDP traffic. Multicast will not affect HTTP traffic (text chat) as that is over TCP.

    3

  • 8/8/2019 Simulation of DoD Collaboration Services

    4/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Text Chat

    Text chat is assumed to be similar to Google Talk and AOL Instant Messenger (AIM). HTTP is

    assumed to be the protocol used to send the chat messages.

    Attributes2

    Page interarrival time (seconds) Exponential(60) Object size (bytes) Uniform integer(100, 1000) Number of objects per page Uniform integer(1, 5)

    Users

    1 500 5000

    Simulated Time

    10 minutesResults

    The simulation produced the results shown below. The response time in the first row of graphs is

    approximately 0.02 seconds, regardless of the user load. The second row of graphs illustrate that

    traffic sent is equal to traffic received which indicates successful delivery of the messages.

    As shown in the third row graphs, the utilization of the DS3 is only 1%. Note, when the usersfirst start sending messages, there is a spike of 3.5% utilization. This is due to the start time of

    2Text chat was replicated using the HTTP. Specifically, the page interarrival time is the chat message frequency.

    The object size is the size of the chat message. Objects per page are the number of chat participants. The main

    difference is the simulation entails a client requesting pages instead of what happens in an actual chat session: the

    client sending chat messages. However since the assumption is a symmetric duplex link, the simulation will provide

    the same usage pattern just in reverse.

    4

  • 8/8/2019 Simulation of DoD Collaboration Services

    5/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    the users set between 100 and 110 seconds in the OPNET. The 100 second lag provides fornetwork convergence. The users all start sending messages at approximately the same time

    initially. Then the exponential interarrival rate of 60 seconds evens out the load. Even in the

    5000 user case, the network is able to support the user load.

    The second set of graphs illustrates the impact of connecting the users through a 56 kbps link. As

    can be seen, response time and network utilization are unacceptable.

    Results for text chat over OC3-DS3: response time (1st row), network traffic (2nd row) and

    network load (3rd row):

    5

  • 8/8/2019 Simulation of DoD Collaboration Services

    6/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Results for text chat over OC3-56k (starting top left, clockwise): response time, traffic

    sent/received, throughput KC POP to STRATCOM, throughput DECC to KC POP:

    6

  • 8/8/2019 Simulation of DoD Collaboration Services

    7/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Audio Collaboration

    Audio collaboration encodes a users voice into digital data and sends over UDP. A single frame

    of audio represents one second of talk from a user and is sent as 1 frame per network packet. Ifthere is silence, a frame does not contain as much data as when the user talks (makes noise).

    Attributes

    Silence length (seconds) Exponential(0.65) Talk length (seconds) Exponential(0.352) Encoder G.723.1 5.3K (silence) Voice (frames/packet) 1

    Users

    1 VoIP-DES-1 50 VoIP-DES-2 100 VoIP-DES-3 500 VoIP-DES-4

    Simulated Time

    5 minutes A short duration is sufficient since audio collaboration session produceconstant network load. Note a longer simulation of a couple users for 1 hour was run toverify the validity of short runs.

    Results

    The simulation produced the results shown below. The response time is consistent for all users at

    around 0.1 second shown in the first row of graphs. As shown in the second row of graphs,traffic sent is equal to traffic received. The third row of graphs show the network is only

    7

  • 8/8/2019 Simulation of DoD Collaboration Services

    8/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    experiencing about 8% utilization for 500 users. The audio collaboration produces a load directlyproportional to the number of users. Response time is steady for all user distributions.

    Shown in the second set of graphs, if a DS1 (1.544 Mb/s) was used instead of the DS3, 500 users

    would saturate the link and cause significant delay. The response time is 0.45 seconds. Data

    received is about one-half the data sent due to the smaller on the DS1 link.

    Also, the network load is not a smooth straight line since silence suppression is used (shown asjagged line in the second row of graphs).

    Results for audio over OC3-DS3: response time (1st row), network traffic (2nd row) and network

    load (3rd row):

    8

  • 8/8/2019 Simulation of DoD Collaboration Services

    9/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Results for audio over OC3-DS1 (starting top left, clockwise): response time, trafficsent/received, throughput KC POP to STRATCOM, throughput DECC to KC POP:

    9

  • 8/8/2019 Simulation of DoD Collaboration Services

    10/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Video CollaborationVideo collaboration encodes a moving image into digital data. The quality of the image is based

    on how many frames (snapshots) of the moving image are encoded per second. For the

    simulation, 5 frames per second were used. The amount of data in each frame is based on howmuch data is needed to encode a moving picture (due to motion or scenery changes). For

    example, a still object requires little data in the frames once the picture is encoded.

    10

  • 8/8/2019 Simulation of DoD Collaboration Services

    11/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Since video collaboration is between people (many times a talking head), there will be quite a bitof movement, some people move more than others. But, there will still be reduced amount of

    data the frames.

    This simulation illustrates a maximum load since the number of bytes for each frame is constant.

    Attributes Frame interarrival time 5 frames/second Frame size 128x120 pixels

    Users

    1 DES-1 10 DES-2 50 DES-3 100 DES-4

    Simulated Time

    5 minutes A short duration is sufficient since audio collaboration session produceconstant network load. Note a longer simulation of a couple users for 1 hour was run toverify the validity of short runs.

    Results

    The first row of graphs shows the response time. As can be seen, the response time degrades(increases) when 100 users are participating compared with 1, 10 and 50. At 50 users, the

    response time is less than 0.02 seconds. But at 100 users, the response time is over 0.2 seconds

    outside the required 0.1 seconds. The degradation is due to the network load produced by 100users saturating the network link.

    The second row of graphs shows the video services sending more traffic than is being received

    due to the congestion. Since the consumers are connected via a DS3, the link is 100% utilized asthe bit rate approaches 45 Mbps as shown in the third row of graphs.

    The second set of graphs show a simulation over a DS1 for 10 users. The response time is at 0.45seconds and the throughput reduced due to the link between the KC POP and STRATCOM only

    capable of handling 1.5 Mbps.

    Note the throughput and utilization graph are smooth. This is due to the constant number of bytes

    in each video frame. As a reference point, a user produces a load of about 700 Kbps. The load islinear due to the constant rates. So, 50 users produce a load of 35 Mbps (80% utilization), shown

    by the light green line in the third row of graphs.

    Results for video over OC3-DS3: response time (1st row), network traffic (2nd row) and network

    load (3rd row):

    11

  • 8/8/2019 Simulation of DoD Collaboration Services

    12/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    12

  • 8/8/2019 Simulation of DoD Collaboration Services

    13/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Results for video over OC3-DS1 (starting top left, clockwise): response time, traffic

    sent/received, throughput KC POP to STRATCOM, throughput DECC to KC POP:

    13

  • 8/8/2019 Simulation of DoD Collaboration Services

    14/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Aggregate

    Simulation the services individually, a DS3 link was able to perform with a response time of 0.1

    seconds or less and a network utilization of 80% or less. However, the services wont necessarilyrun separately and it is important to simulate the aggregate load.

    Attributes

    Text chat Same as individual simulation (above) Audio Same as individual simulation (above) Video Same as individual simulation (above)

    Users

    Text Audio Video Simulation Name

    1 1 1 DES-1

    500 100 10 DES-2

    5000 500 50 DES-3

    Simulated Time

    10 minutesResults

    The simulation produced the results shown below. When aggregated, the network is unable tohandle the user load that was handled in the single service cases. Specifically, when text, audio

    and video are at 5000, 500 and 50 users, respectively, the network is incapable of operating

    within 0.1 seconds response time as shown in the first row of graphs.

    14

  • 8/8/2019 Simulation of DoD Collaboration Services

    15/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    15

    The cause of the delay is mainly due to the video load as shown in the second row of graphs. Thevideo traffic sent is much more than the traffic received. Whereas, the text and audio services are

    close to equal although slightly degraded. However, unlike the video single service case above

    the utilization is not at 100% as shown in the third row of graphs. This is due to the fact that therouters at the DECC, KC POP and STRATCOM drop packets as shown in the third graph in the

    third row.

  • 8/8/2019 Simulation of DoD Collaboration Services

    16/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Results for aggregate over DS3: response time (1st

    row), network load (2nd

    and 3rd

    rows):

  • 8/8/2019 Simulation of DoD Collaboration Services

    17/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

  • 8/8/2019 Simulation of DoD Collaboration Services

    18/18

    Modeling and Simulation Final Project SYS 611 Modeling and Simulation

    Conclusion

    Three collaboration services were simulated: video, audio and text chat. By simulating each

    service individually and varying the number of users, it was shown that the network utilization

    increases linear based on the number of users. The response time remains constant as long as the

    network load does not approach 100% utilization. At 100% utilization, the network fails todeliver the service to the users in a timely manner.

    As shown in the table below for the individual simulations, the maximum number of simulated

    users were supported with the following response times, network throughput and utilization.

    Service

    Type

    Users Response Time

    (seconds)

    Throughput

    (megabits/second)

    Utilization (%)

    Text 5,000 0.02 0.4 1

    Audio 500 0.1 3.5 8

    Video 50 0.02 35 80

    The aggregate simulation illustrated that none of the services operated within the requiredresponse time as shown in the table below.

    Service

    Type

    Users Response Time

    (seconds)

    Throughput

    (megabits/second)

    Utilization (%)

    Text 5,000 35

    Audio 500 4

    Video 50 4

    Total 13 30

    The recommendation is the DoD not proceed with the full deployment of the collaboration

    services at this time. There is too much risk that the network will not be able to support the user

    load. Instead, the DoD should pilot the collaboration services at 10-20% of the user load togather empirical data on the network characteristics. It was shown that 500 text users, 50 audio

    users and 10 video users running concurrently operated within the performance parameters. Alsothe simulation did not model any background network traffic which will have an affect on the

    network load. By deploying a scaled down version of the services, the DoD can get a better

    understanding of the impact to the network and promote a more successful deployment in the

    long run.

    18