Generic selectors
Exact matches only
Search in title
Search in content
Search in posts
Search in pages

Packet-switched data

Viewing 15 posts - 1 through 15 (of 25 total)
  • Author
    Posts
  • #32374 Reply
    Jyrki Laine
    Guest

    Before the actual question: If I have understood we´ll calculate the needed trunks for the speech traffic with Erlang B formula. If we know the speech traffic per subscriber during the busy hour (e.g. 0.03 Erl), blocking should be less than 2% and there are e.g. 100 000 subscribers.

    But how shall we calculate the real traffic situation if we have also e.g. 50 000 packet switched data subscribers and they`ll transfer during the busy hour e.g. 3000 kbit/s of packet-switched data per subscriber. And the average user data rate should be 128 kbit/s (144kbit/s).

    Please answer to this question. Do I need any other data for calculation?

    Best regards,
    Jyrki Laine

    #32375 Reply
    Jyrki Laine
    Guest

    I`m sorry but there is one mistake in my article: I meant that everyone of the 50 000 subscribers will transfer 3000 kbit (NOT 3000kbit/s) of packet switched data during the busy hour.

    Best regards,
    Jyrki Laine.

    #32376 Reply
    spring
    Guest

    Hello Jyrki
    Please forward me all respond on your question.I have very similar problem.
    Thank you kindly.
    Pls send mail to xcguo@163.net

    #32377 Reply
    Milena
    Guest

    Hi Jyrki,

    I read your question about the packet-switched data capacity planning. I have the same problem and haven’t been able to find a solution yet. Have you found an answer yet?

    I would greatly appreciate any ideas, references, etc.

    Thanks,

    Milena (milena@glue.umd.edu)

    #32378 Reply
    lucio
    Guest

    Got a similar problem! Please help me!

    Thanks
    Lucio

    #32379 Reply
    Nicolae Panait
    Guest

    Hy everybody,

    This topic is very complex. I haven’t found a solution neither, but I think we should consider different busy hours (in fact periods- more then one hour), different GoSs and different bandwith requirements for every data service we offer.

    I know is not an easy way to do that. I’m still working at this. Until we can conclude with a magic way of doing this, here is what I found very interesting:

    http://www.cstp.umkc.edu/personal/dmedhi/abs/mg_tr_6_r1_95_abs.html.
    [Webmaster’s note: requires a PostScript viewer.]

    Hope this helps,
    Nicolae

    #32380 Reply
    Akik
    Guest

    Hi guys,

    In order to calculate the total data traffic that needs to be supported, one can take following steps.

    1. First divide 3000kbits per hour by the user data rate of 128 kbps. (3000kpbper busy hour/128kbps)*(3600s/1 hr) = 6.5 mErlang per subsciber.
    2. Now 6.5 mErlang/sub * 50000 subscirber will give you total of 325.52 Erlang of data traffic that the network needs to support.

    Hope you found this helpful….

    #32381 Reply
    Westbay Engineers Ltd.
    Guest

    We studied the article suggested by Nicolae Panait. But it does not consider data traffic. It is about voice, although it does discuss multi-service (eg. 64kbps voice and 384kbps video together).

    It seems to me that Erlang B, which uses blocking to define it’s grade of service just won’t work for data. Delay should be the metric used for data. So, perhaps Erlang C would be more appropriate.

    Any thoughts?

    #32382 Reply
    Rommel
    Guest

    You are designing your backbone connection between two routers. You know that you will generally see about 600 packets per second and 200 bytes per packet or 1600 bits per packet. The result of multiplying 600 pps by 1600 bits per packet gives us the amount of bandwidth we will need to support, 960,000 bps. We know that we can buy circuits in increments of 64,000 bps. This will give us the amount of data necessary to keep the circuit busy for 1 second. How many circuits will we need to keep delay under 10 milliseconds?
    To get the traffic load we will need to:
    960,000 bps/64,000 bps = 15 Erlangs of traffic load
    To get the average transmission time we will need to multipy bytes per packet by 8 to get bits per packet. Then divide by 64000 bps (circuit speed) to get the average transmission time per packet
    200 bytes/packet * 8 bits = 1600 bits/packet /64000 bps = 0.025 seconds to transmit or 25 milliseconds.
    Delay factor 10 ms/25 ms = 0.4 delay factor

    By using an erlang C table you will get a result of ~17 64K circuits

    #32383 Reply
    Rommel Bajamundi
    Guest

    Definitely Erlang C should be used. Here is an example of how to use it:

    You are designing your backbone connection between two routers. You know that you will generally see about 600 packets per second and 200 bytes per packet or 1600 bits per packet. The result of multiplying 600 pps by 1600 bits per packet gives us the amount of bandwidth we will need to support, 960,000 bps. We know that we can buy circuits in increments of 64,000 bps. This will give us the amount of data necessary to keep the circuit busy for 1 second. How many circuits will we need to keep delay under 10 milliseconds?
    To get the traffic load we will need to:
    960,000 bps/64,000 bps = 15 Erlangs of traffic load
    To get the average transmission time we will need to multipy bytes per packet by 8 to get bits per packet. Then divide by 64000 bps (circuit speed) to get the average transmission time per packet
    200 bytes/packet * 8 bits = 1600 bits/packet /64000 bps = 0.025 seconds to transmit or 25 milliseconds.
    Delay factor 10 ms/25 ms = 0.4 delay factor
    25ms = AHT
    10ms = avg delay
    When I look in my erlang tables, I see that 0.4 blocking probability or delay factor with 15.25 Erlangs of traffic load yields, 18 Agents (traffic sensitive facilities)

    In this context an erlang is used as the time it takes to keep a traffic sensitive facility busy for 1 second.

    #32384 Reply
    Richard Arcadi
    Guest

    Hi to all:

    Maybe I’ve not understood the basis of the problem. Are we talking about circuit switched or packet switched networks?

    From my own expererience, I’ve never mixed Erlang circuit) with Poisson (packet).

    Even when a switched circuit is being used, the transfer of packetized data within it is far better modelled by Poisson and Markov models of “packet store and forward links”.

    Poisson-Markov allows to contemplate the queuing of packets in a given node, related to the capability of the node to “tose out” the packets of the incoming link and is highly sensitive to the packet lenght.

    I’d use Poisson to model one circuit seizure, and then I’d use erlang.

    rarcadi@hotmail.com

    #32385 Reply
    Chris Taylor
    Guest

    I think this thread is about packet switched (data, not voice).

    I’ve never heard about the Poisson model being used in this situation. As far as I knew, Poisson is very similar to Erlang B, except in the way that blocked calls are considered (ie. they remain in the system).

    But, with packet switched data, there is no blocking.

    #32386 Reply
    Rommel
    Guest

    Richard,
    My understanding is that the Markov Modulated Poisson Process(MMPP) model is used for traffic shaping…not traffic analysis. MMPP takes into account the peakedness of data traffic and the effect of aggregated traffic flows into a single pipe. Simply put MMPP model gives us a method in which we drop packets in order to avoid congestion in a circuit.
    It is a great model to use to create an algorithm like Random Early Detection (RED) but nothing in the model tells me how “fat” my pipe should be.

    #32387 Reply
    Richard Arcadi
    Guest

    Sorry for any misconception I could introduce with my former posting. I wanted to remark the fact that the randomness of data network behavior has been explained in a more simplier way with these facts:

    1. Poisson describes input and output traffic behavior.
    2. Markov describes (Model M/M/1) the behavior of the output route manager (1 stands for one I/O).
    3. There are parameters to predict the average behavior of the router, like: packet sizes, queue depth, binary speed and so.

    There is no circuit seizure, as the model operates with packet store and forward. There is that I didn’t understand the use of Erlangs.

    #32388 Reply
    Jose
    Guest

    Hi everybody
    I´m working in my proyect in my university.
    I need how to estimate the bandwitdh of the backbone of a network for voice, data and video. Here you give a lot of information about to calculate the trunks for voice, even video. But how can I estimate the bandwidth for data? Do you have any example? There are usual ratios for typical data aplications?

    thanks,
    email: perenquen@hotmail.com

Viewing 15 posts - 1 through 15 (of 25 total)
Reply To: Packet-switched data
Your information:




<a href="" title="" rel="" target=""> <blockquote cite=""> <code> <pre class=""> <em> <strong> <del datetime="" cite=""> <ins datetime="" cite=""> <ul> <ol start=""> <li> <img src="" border="" alt="" height="" width="">