Traffic Engineering (TE)

Document Sample
Traffic Engineering (TE) Powered By Docstoc
					                                                                                           Traffic Engineering (TE)
                                                                                            Dynamically reconfiguring the network topologies or routes so as
                                                                                            to effectively accommodate current traffic
                   Estimating current traffic matrices                                         Example:Optical layer Traffic Engineering (TE)
                                                                                                  Environment: IP-over-WDM networks
                   accurately by using long-term                                                                      p
                                                                                                          Construct optical p           physical network
                                                                                                                             paths over p y
                   variations information                                                                 A set of optical paths forms a virtual network topology for IP
                                                                                                          network
                                                                                                   Method:
                                                                                                          Reconfigure the optical paths according to the current traffic
                         Yuichi Ohsita
                         Osaka Univ.

                                      Broadnets 2008                                   1                                      Broadnets 2008                                2




Input of Traffic Engineering Methods
                                                                                           Traffic Matrix Estimation
Current traffic demands between all nodes (traffic matrices)                               Traditional estimation methods:
                                                                                              Estimate from the equations
  Hard to monitor directly                                                                    based on the link loads
        Construct fully meshed label switched paths (LSPs)
                                                                                                 X = AT
             Require N-squared LSPs                                                              X: Link loads
        Count packets of each flow                                                               A: Routing matrix
             Require non-negligible amount of CPU resources at the edge nodes                    T: Traffic matrix                                Link load is sum of the
                                                                                              Estimated traffic matrices include                  amounts of traffic traversing
                                                                                              estimated errors                                    the link
   Estimate traffic matrices from link loads                                                     The number of equations is much smaller than the number
        Link loads are monitored more easily even in large network                               of entries in traffic matrix


                                                                                              The estimation errors degrade the performance of
                                                                                              traffic engineering

                                      Broadnets 2008                                   3                                      Broadnets 2008                                4




Using the Additional Information                                                           Necessity of Consideration of Traffic Variations
   Method to increase the accuracy of estimation by using
   additional measurements                                                                   If it takes a long time to obtain sufficient number
       The additional information is obtained by using the                                   of measurements…
       measurements both before and after traffic engineering [12]
                                                                                               Current traffic may differ from the initially monitored
                                                                                               traffic
     TM Estimation             Estimated TM            Traffic Engineering
                                                                                                    We need to consider the traffic variations
                     Additional Measurements                                                      Traffic
                                                                                                  Volume


⎡ X (n − M + 1) ⎤ ⎡ A(n − M + 1)⎤
⎢                                 ⎥T X (n) : Link to the equations
        Add the link loads monitored at each time loads monitored at time n
         M      ⎥=⎢        M
                                                                                                                            ・・・・・・・・・・・・
⎢               ⎥ ⎢               ⎥ A (n ) : Routing matrix at time n
⎢
⎣      X (n)    ⎥ ⎢ A(n) ⎥
                ⎦ ⎣               ⎦                                                                                                                                        Time
                                                                                             Monitor
   [12] Y Ohsita et al, “Gradually Reconfiguring Virtual Network Topologies based on         link loads
   Estimated Traffic Matrices,” in Proc. INFOCOM 2007, May 2007                        5                                      Broadnets 2008                                6




                                                                                                                                                                                  1
  Goal of this work                                                                           Overview of proposed method
                                                                                                                                       (1) Estimate long-term variations
  Estimate traffic matrices accurately by collaborating                                                                                   ・Model the variations as periodical functions
                                                                                                                                             ・Estimate the parameters in the model
  with traffic engineering                                                                                                                   by using the both previously and currently monitored
                                                                                                                                             link loads




                                                                                              betw
                                                                                              Traff Volume
      Obtain the additional information by using the route




                                                                                                 ween OD pair
                                                                                                  fic
       h             d by traffic    i
      changes caused b t ffi engineering  i
      Use the additional information considering the traffic
      variations

                                                                                                                                                                                           Time
                                                                                            (2) Adjust the estimated                                         (3)If the estimated long-term variation
                                                                                            variations so as to fit the                                      does not match the current variations
                                                                                            current link loads                                               ・Detect the mismatch
                                                                                                                                                             ・Re-estimate the long-term variations
                                        Broadnets 2008                                  7                                                               Broadnets 2008                                8




  Estimation of Long-term Variations                                                          Adjustment of the Estimated Long-term Variations

     We model the long-term variations as periodical                                                  Obtain the current traffic matrices which fits the
     functions [14]                                                                                   current link loads and is close to the estimated
           By using the Fourier series expansion, the periodic                                        long-term variations
           functions are represented as

                                                                                                                                                                                g       g
                                                                                                                                                                            Range fitting the
                                                                                                                                       Estimated long-term variations       current link loads
Traffic between nodes i and j at time n                  Cycle
           We estimate by setting       so as to fit all the link                                                                                                                        Estimated
           loads (including previously monitored link loads)                                    s.t.                                                                                     current traffic
                                                                                                                                                                                         matrix

                                                          where
                                                                                                                                                  Estimated long-term
                                                                                                                                                  variations
     [14] A. Soule, A. Nucci, R. Cruz, E. Leonardi, and N. Taft, “Estimating dynamic
     traffic matrices by using viable routing changes,” IEEE/ACM Transactions on
                                                                                        9                                                               Broadnets 2008                               10
     Networking, vol. 13, pp. 485–498, June 2007.




  Handling the Change of Variations                                                           Detection of the Change of Variations
        When traffic variations change                                                                Detect the change when | tiestj (n) − tˆi , j (n) | is
                                                                                                                                   ,

                                               Traffic changes                                        significantly larger than before.
                                                significantly
                                                                                                                The traffic matrices after adjustment fit the current
                                                                                                                traffic
                                                                                                                       If the estimated long-term variations is far from the current
                                                                                                                       traffic | tiest (n) − ti , j (n) | is large
                                                                                                                       traffic,     ,j
                                                                                                                                             ˆ
             Monitored link loads in this period are far from the current traffic
             variations                                                                                         The detection is done by Smirnov-Grubbs tests
            If we use the link loads monitored before the change, we cannot
                        long-
             estimate long-term variations so as to fit the current traffic                                                                            Detect the
                                                                                                                | tiest (n) − ti , j (n) |
                                                                                                                    ,j
                                                                                                                              ˆ
                                                                                                                                                        change
        Steps to handle the change of variation
      1.     Detect the change
      2.     Delete information before the change
      3.     Re-estimate the long-term variations
                                        Broadnets 2008                                 11                                                                                                            12




                                                                                                                                                                                                           2
Deletion of Information before the Change                                       Evaluation
  Delete only information corresponding to
  the changing traffic                                                            Environment
     From routing matrices                                                          Topology
                                                                                       EON backbone topology
           A(i)    Set elements corresponding
                                                       A' (i )
                    to the detected traffic to 0                                    Traffic demand:
                                                                                       Generate by adding variations to sin functions
     From monitored link loads                                                         whose amplitudes and phases were randomly
           Remove the estimated amount of the detected                                 generated.
           traffic from each link loads                                             Traffic engineering method:
                                                         Estimated amount of           Optical layer TE
                                                         the detected traffic               Add optical layer paths so as to make the maximum link
                                                                                            utilization less than 0.7
  Re-estimate long-term variations by using                                            We perform the TE method once an hour
  A' (i ) and X ' (i )
                                                                         13                                   Broadnets 2008                         14




Accuracy of the estimation                                                      Estimated Amount of Traffic Changing Significantly
 Methods
   Tomogravity method
      Method using only the                                                       Method
      currently monitored link loads                                                Our method (with re-estimation)
   Additional Equation method                                                       Our method (without re-estimation)
      Method using link loads
      monitored at previous times but                                             Results
        t     id i the traffic    i ti
      not considering th t ffi variations                                           By re-estimating the long-term variations, we can estimate the
   Our method (without re-estimation)                                               traffic amount even when traffic changes significantly
   Our method (with re-estimation)
 Metrics
                                          ∑(estimatedvalue− actualvalue)
                         1                                                2
   (RMSE) =                                                                          With
                  numberof elements                                                                                            Without
 Results                                                                             re-estimation                             re-estimation
   Our method can estimate traffic matrices the most accurately
      Our method can use many additional information considering the
      traffic variations                                                 15                                                                          16
                                 Broadnets 2008                                                               Broadnets 2008




Maximum link utilization achieved by TE
                                                                                Conclusion
using estimated traffic matrices
 Methods
   Tomogravity method
                                                                                  We propose an estimation method which uses
      Method using only the
                                                                                  additional measurements considering the traffic
      currently monitored link loads                                              variations
   Additional Equation method                                                       Estimate long-term variations by using the link
      Method using also link loads                                                  loads monitored at previous times
      monitored at previous times but                                                                     long term
                                                                                    Adjust the estimated long-term variations so as to fit
      not considering the traffic variations                                        the current link loads
   Our method (without re-estimation)                                               Re-estimate the long-term variations, when the traffic
   Our method (with re-estimation)                                                  variations changes significantly
 Metrics                                                                          Evaluation results show that our method can
   Maximum link utilization after TE performed by using the                       estimate traffic matrices accurately.
                               target value=0.7)
   estimated traffic matrices (target value=0.7
 Result
   Our method (with re-estimation) can decrease the maximum link
                            Broadnets 2008                               17                                   Broadnets 2008                         18
   utilization sufficiently




                                                                                                                                                          3