Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 13 views

    Poisson Processes, Markov Chains and M/M/1 Queues: Advanced Communication Networks

    Uploaded by

    Ashish Kumar
    The document discusses queueing models such as the M/M/1 and M/M/m models. The M/M/1 model describes a queue with Poisson arrivals and exponential service times with one server. It can be analyzed as a discrete-time Markov chain. The M/M/m model generalizes this to have m servers. Both models can be used to analyze metrics like average queue length and delay. The PASTA property shows that the steady-state distribution seen by arriving customers is the same as the overall steady-state distribution for these models.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd

    Poisson Processes, Markov Chains and M/M/1 Queues: Advanced Communication Networks

    Uploaded by

    Ashish Kumar
    13 views17 pages
    The document discusses queueing models such as the M/M/1 and M/M/m models. The M/M/1 model describes a queue with Poisson arrivals and exponential service times with one server. It can be analyzed as a discrete-time Markov chain. The M/M/m model generalizes this to have m servers. Both models can be used to analyze metrics like average queue length and delay. The PASTA property shows that the steady-state distribution seen by arriving customers is the same as the overall steady-state distribution for these models.

    Original Description:

    cc

    Original Title

    lec5

    Copyright

    © Attribution Non-Commercial (BY-NC)

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document discusses queueing models such as the M/M/1 and M/M/m models. The M/M/1 model describes a queue with Poisson arrivals and exponential service times with one server. It can be analyzed as a discrete-time Markov chain. The M/M/m model generalizes this to have m servers. Both models can be used to analyze metrics like average queue length and delay. The PASTA property shows that the steady-state distribution seen by arriving customers is the same as the overall steady-state distribution for these models.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    13 views17 pages

    Poisson Processes, Markov Chains and M/M/1 Queues: Advanced Communication Networks

    Uploaded by

    Ashish Kumar
    The document discusses queueing models such as the M/M/1 and M/M/m models. The M/M/1 model describes a queue with Poisson arrivals and exponential service times with one server. It can be analyzed as a discrete-time Markov chain. The M/M/m model generalizes this to have m servers. Both models can be used to analyze metrics like average queue length and delay. The PASTA property shows that the steady-state distribution seen by arriving customers is the same as the overall steady-state distribution for these models.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 17

    Review

    Examples

    M/M/m

    Poisson processes, Markov chains and M/M/1 queues

    Advanced Communication Networks

    Lecture 5

    Review

    Examples

    M/M/m

    M/M/1 Analysis

    Poisson Arrivals, Exponential service times, 1 server (FIFO)

    Look at discrete times With high prob., only one arrival or departure Discrete-time Markov chain
    1 0 1 2 3 1 1 1

    Review

    Examples

    M/M/m

    Analysis Contd..
    Steady state prob.s {pn } Balance equations pn = pn1 pn = pn1 =
    n

    p0 = n p0

    pn = 1 p0 = 1
    n

    pn = (1 )n
    Stability of system: <

    Review

    Examples

    M/M/m

    Computing system averages


    Comment: pn = f ( ) As 0 same answers for cont. time model Average number in system

    N=
    n =0

    npn =
    n =0

    nn (1 )

    = = 1 Average system delay: Littles law T = As 1, N , T N 1 =

    Review

    Examples

    M/M/m

    100 90 80 70 60

    50 40 30 20 10 0 0 0.2 0.4 0.6 0.8 1

    Single pole response - typical of queuing systems Other system variables W =T


    1

    NQ = W =

    2 1

    Review

    Examples

    M/M/m

    Compare to D/D/1 queue


    one arrival every
    NQ

    sec, 1 departure every


    D/D/1

    sec

    =1

    NQ = 0 < 1. M/M/1: queue size, delay blows up for near 1 Intuition: Variability causes performance loss

    Review

    Examples

    M/M/m

    Changing transmission rate

    M/M/1 queue: Arrival rate and Service time is doubled What happens to delay? N?

    2, 2 stays same N= N stays same 1 T = 1 = 1


    1

    T reduces by

    1 2

    Review

    Examples

    M/M/m

    T old

    T new

    Review

    Examples

    M/M/m

    Example 2
    Statistical Multiplexing
    /M

    TDM
    /M

    /M

    /M

    /M

    /M

    M independent Poisson streams, rate TSM = 1

    TTDM =

    Delay reduced by factor of m Statistical Multiplexing gain


    Cons: Difcult to isolate Bad ows ; Provide guarantees

    Review

    Examples

    M/M/m

    Distribution of System Variables

    What about distribution for N? For eg: Variance

    Var (N ) =
    n =0

    n (1 ) n2 (1 )2

    Likewise, Distn. for T in Prob. 3.1 (M/M/1 queues)

    Review

    Examples

    M/M/m

    PASTA property
    Interested in state of system just before packet arrives Eg: to calculate Blocking probability Steady-state prob. arriving packet sees in the system
    t

    lim Pr( N (t ) = n| arrival @t + ) ?

    Isnt this same as pn ? Not necessary Eg: D/D/1 system, < Above prob. is zero, but no steady-state exists

    Review

    Examples

    M/M/m

    PASTA property Contd..


    For Poisson trafc,

    The two quantities are equal PASTA Proof an = P{ N (t ) = n| arrival @t + } = P{ N (t ) = n| A(t , t + )} P ( A(t , t + )| N (t ) = n) P (N (t ) = n) (t )=n,A(t ,t + )) = P( N P = (A(t ,t + )) P (A(t , t + ))
    But A(t , t + ) ind. of N (t ) = n an (t ) = P (N (t ) = n) Holds for broad class of queuing systems w/ Poisson arrivals, ind service distribution (Memoryless property)

    Review

    Examples

    M/M/m

    M/M/1 - Last slide

    What is the prob arriving customer nds system empty? p0 = 1

    Review

    Examples

    M/M/m

    Other Markovian systems: M/M/m


    0 1 2 2 3 3 m m+1

    m servers in system
    Given 2 packets in service, Prob of departure = Prob(1st packet departs) + Prob(2nd packet departs)

    = + Service time of state n = m n > m


    Ex: Circuit switched networks, blocked calls wait

    Review

    Examples

    M/M/m

    M/M/m analysis

    Balance equations

    pn1 ( ) = pn (n ) n = 1, 2, , m pn1 ( ) = pn (m ) n > m pn = pn = n (n 1) m


    n m

    p0
    m

    nm n>m

    1 m!

    p0

    Review

    Examples

    M/M/m

    M/M/m analysis Contd..

    pn =
    If stable: pn = 1 Stability condition: =
    n =0

    1 n n! (m) p0 m m n m!

    nm n > m,

    <1

    m 1

    p0 =
    n =0

    (m)n (m)m + n! m!(1 )

    Review

    Examples

    M/M/m

    Erlang C formula
    What is the prob arriving packet has to wait for service? Same as prob that servers are busy (Follows from PASTA property)

    PQ = Pr(Queuing) = Pr(N m)

    =
    n =m

    pn = p0

    (m)m m!(1 )

    Widely used in telephony Model for Blocked delay calls Formula also hold for M/G/1 systems (Invariance property)

    You might also like