manish by gegeshandong

VIEWS: 8 PAGES: 32

									Available Bandwidth
     Estimation

                   Manish Jain
 Networking and Telecom Group
             CoC, Georgia Tech
                    Outline
 Introduction and definitions
 Estimation methodologies
      Train of Packet Pairs(TOPP)
      Self Loading Periodic Streams (SLoPS)
      Packet Train Gap Model

   Open Issues


09/23/2003          8803 Class Presentation    2
                                Definition
    Available Bandwidth: unutilized capacity



   Varies with time
   ui : utilization of link i in time interval t ( 0 <= ui <= 1 )
   Available bandwidth in link i:
                        Ai  Ci(1 - ui)
   Available bandwidth in path (Avail-bw):
                    A  min Ai  min Ci (1 - ui)
                                i 0..H           i 0..H
                Tight link: minimum avail-bw link


09/23/2003                                8803 Class Presentation   3
 Available Bandwidth:time varying
              metric t


             A(t)




                               T


                                t

   t defines sampling/averaging timescale
   Average avail-bw in t
      Does not tell how avail-bw varies
      Variation range gives more information



09/23/2003               8803 Class Presentation   4
             Why do we care ?
 ssthresh in TCP
 Streaming applications
 SLA verification
 Overlay routing
 End-to-end admission control




09/23/2003        8803 Class Presentation   5
        Measuring per-hop available
                bandwidth
   Can be measured at each link from interface
    utilization data using SNMP
   MRTG graphs: 5-minute averages




   But users do not normally have access to SNMP data
   And MRTG graphs give only per-hop avail-bandwidth

09/23/2003             8803 Class Presentation           6
                 Measuring path Available
                       Bandwidth
   Blast path with UDP packets
        Intrusive
        Carter & Crovella: cprobe (Infocom 1996)
                Packet train dispersion does not measure available bandwidth
                 (Dovrolis et.al. Infocom’01)
   Measure throughput of large TCP transfer
        TCP throughput depends on network buffer
   Ribeiro et.al. : Delphi (ITC’00)
        Correct estimation when queuing occurs only at single link
        Assumes that cross traffic can be modelled by MWM model




09/23/2003                         8803 Class Presentation                      7
A New End-to-end probing and
     analysis method for
    estimating bandwidth
         bottlenecks

B. Melander et al, In Global Internet
Symposium, 2000




                                        8
                      Introduction
     In FCFS queue, output rate is function of
      input rate and cross-traffic rate
                    Oj-1                         Oj                                Oj+1
                           Cj                                    Cj+1
                                                                                                     Cj+1-Mj > Cj-Mj-1
             Mj-1                   Mj
                                 
   In one hop:                  o
                           o j   j 1 j 1
                                        o
                                  o j 1mj 1*C j
                                                                         if o j 1C j m j
                                                                          if o j 1C j -mj
                                 

                                     o j 1                                  if o j C j 1 m j 1&o j 1C j m j
                                     o

    In two hop:
                                             j 1
                                                          *C j                if o j C j 1 m j 1&o j 1 C j m j
                                    o j 1  m j 1
                                    
                           o j 1               o j 1
                                            o j 1m j 1
                                                                 *C j *C j 1 if o j C j 1-m j 1&o j 1 C j m j
                                     o
                                              j 1
                                                         m j 1
                                     o j 1  m j 1
                                    



09/23/2003                   8803 Class Presentation                                                                     9
                           Key Idea:TOPP
        o :sending rate
        f: receiving rate
                                                                Break points
             o
                a i  bio
             f
                where i is number links
                 with different available
                 bandwidth
        For i=1
                b1=1/Ctight
                a1=1-Atight/Ctight



09/23/2003                            8803 Class Presentation                  10
                               Algorithm
   Algorithm:
                Send n probe pairs with a minimum rate
                Record receive rate at receiver
                Increment rate by fixed d and repeat
                Measure available bandwidth from the relation of o/f vs o
   Avail-bw and capacity of other links can be
    measured
      if        links in ascending order of avail-bw
   In practice, break points may be hard to
    identify

09/23/2003                        8803 Class Presentation               11
End-to-end Available Bandwidth:
  Measurement Methodology,
Dynamics and Relation with TCP
         Throughput

 M. Jain and C. Dovrolis, In IEEE/ACM
 TON, August 2003




                                12
                          Key idea: SLoPS


   Examine One-Way Delay (OWD) variations of a fixed rate
    stream
        Relate rate to avail-bw
                                    R
    OWD: Di = Tarrive-Tsend = Tarrive - Tsend + Clock_Offset(S,R)
                      R S                     R

   SLoPS uses relative OWDs, DDi = Di+1 – Di-1 (independent of
    clock offset)
   With a stationary & fluid model for the cross traffic, and FIFO
    queues:
                If R > min Ai, then DDi > 0 for I = 1…N
                Else DDi = 0 for for I = 1…N


09/23/2003                              8803 Class Presentation   13
             Illustration of SLoPS
   Periodic Stream: K packets, size L bytes, rate R = L/T




   If R>A, OWDs gradually increase due to self-loading of stream



09/23/2003                  8803 Class Presentation                 14
               Trend in real data




   For some rate R
      Increasing trend in OWDs  R > Avail-bw
      No trend in OWDs  R < Avail-bw




09/23/2003                  8803 Class Presentation   15
    Iterative algorithm in SLoPS
   At sender: Send periodic stream n with rate Rn
   At receiver: Measure OWDs Di for i=1…K
   At receiver: Notify sender of trend in OWDs
   At sender: If trend is :-
            increasing (i.e. Rn >A )  repeat with Rn+1 < Rn
            non-increasing (i.e. Rn <A )  repeat with Rn+1>Rn
                Selection of Rn+1 : Rate adjustment algorithm
   Terminate if Rn+1 – Rn < 
    : resolution of final estimate



09/23/2003                          8803 Class Presentation       16
    If things were black and white…




   Grey region: Rate R not clearly greater or smaller
    than Avail-bw during the duration of stream
        Rate R is within variation range of avail-bw



09/23/2003                    8803 Class Presentation    17
                 Big Picture




   Increasing trend  R > variation range of
    Avail-bw
   No trend  R < variation range of Avail-bw
   Grey trend  R inside variation range

09/23/2003          8803 Class Presentation      18
                        Rate adjustment algorithm
                                                            Increasing trend :
                                           Rmax   >A           Rmax = R(n)
Variation Range




                                           Gmax                R(n+1) = (Gmax + Rmax)/2
                          Grey region                       Non-increasing trend:
                                           Gmin
                                                               Rmin = R(n)
                                                               R(n+1) = (Gmax +Rmin)/2
                                           Rmin < A         Grey region & R(n) > Gmax:
                                                               Gmax = R(n)
                  Terminate if:                                R(n+1) = (Gmax + Rmax )/2
                  (Rmax – Gmax) && (Rmin– Gmin) <          Grey region & R(n) < Gmin:
                                                               Gmin = R(n)
                                                               R(n+1) = (Gmin + Rmin )/2



                  09/23/2003                   8803 Class Presentation                     19
How do we detect an increasing trend?
                           (D  D )
                                                                K

     
             K
       I (D  D )                                               j 2
                                                                       j       j  1

                      R 
                    j   j  1
             j 2
 R 
                           |D D |
                                                          pdt
    pct                                                         K
       K 1                                                     j 2
                                                                           j   j  1




 Infer increasing trend when PCT or PDT trend  1.0

09/23/2003                      8803 Class Presentation                                20
             Verification approach
   Simulation
        Multi-hop topology
        Cross traffic: Exponential and Pareto interarrivals
        Varying load conditions
   Experiment
        Paths from U-Delaware to Greek universities and U-Oregon
        MRTG graphs for most heavily used links in path
        Compare pathload measurements with avail-bw from MRTG
         graph of tight link
        In 5-min interval, pathload runs W times, each for qi secs 5-
         min average avail-bw R reported by pathload:
                            W
                                  qi Rimax  Rimin
                       R
                            i 1 300       2

09/23/2003                   8803 Class Presentation                21
             Verification: Simulation
   Effect of tight link load
      Pathload range versus avail-bw during simulation (average of 50 runs)
      5 Hop, Ctight=10Mbps, utilnon-tight=.6 %




   Center of pathload range: good estimate of average of avail-bw

09/23/2003                    8803 Class Presentation                    22
             Verification: Experiment
       Tight link: U-Ioannina to AUTH (C=8.2Mbps), =1Mbps




09/23/2003                  8803 Class Presentation           23
             Avail-bw Variability versus
                   stream length
   Relative variation index:




   Longer probing stream observe lower variability
        However, longer streams can be more intrusive

09/23/2003                     8803 Class Presentation   24
             Avail-bw variability versus
                    traffic load




   Heavier link utilization leads to higher avail-bw variability



09/23/2003                   8803 Class Presentation                25
  Evaluation and
Characterization of
Available Bandwidth
    Techniques
 N. Hu et al, JSAC, August 2003




                            26
    Packet Pair Model: Single Hop
   In single hop path
                Competing traffic may be inserted between packet pair
                Packet pair gap at receiver is function of cross traffic
                                    Gi

    Input
                                         Go                   q        t
    Case1: Go = Gi – q/C < Gi

                                         Go
                                                                        t
                                          m/C
    Case2: Go=m/C+Gb

                                                                        t
    Assumption: Fluid cross traffic
    In practice, CT is bursty
         Packet train will capture average

09/23/2003                          8803 Class Presentation                 27
    Packet Train Model: Single Hop
                                       Gi



                                            Gb           Gi+                           t



                 M                                                                     t
           C *  (g  gB )  
                            i
                 i 1
     M                  K       N                Where Total numer of probing packets = M+K+N

     g  g  g
    i 1
             
             i
                     i 1
                            
                            i
                                i 1
                                        
                                        i


   Assumption:
           Only increased gap sees CT
           Packet dispersion not affected by CT at post-tight link


09/23/2003                                       8803 Class Presentation                        28
             IGI and PTR Algorithm
 Start by sending out packet train with
  minimum gap ( gB)
 If gap@receiver != gap@sender
      Send    another train with increased gap
   Else calculate available bandwidth
      IGI: Use equation
      PTR: Available Bandwidth = Rate of last
       train measured at receiver

09/23/2003             8803 Class Presentation    29
     Summary: Single Hop Model
   IGI:
      Need to know the capacity of tight link
      Assume that tight link is same as narrow link

   PTR:
      Same   as TOPP
   Relation of amount of cross-traffic and
    dispersion
      May   not hold in multi-hop path


09/23/2003               8803 Class Presentation       30
                           Open Issues
   Integrate avail-bw estimation methodology with
    application
                Use data packets in place of probe packets
   Implement avail-bw estimation algorithm in network
    interface card
                Allow routers to do avail-bw estimation
   Can we make some short-term predictions of avail-
    bw?
   High bandwidth paths
                Time stamping packets
                MTU limitations


09/23/2003                        8803 Class Presentation     31
                                  Pathchirp
   Uses exponentially spaced packet train
   Main idea:
                Avail-bw > Rk , if qk >= qk+1
                Avail-bw < Rk , otherwise
                      Can be used when probe packets are close enough
                Identify excursions: consecutive packets show increased
                 queuing delays




                Per-packet avail-bw Ek
                Final estimate: Expected value of Rk


09/23/2003                           8803 Class Presentation               32

								
To top