Traffic Engineering with Traditional IP Routing Protocols by hcj


									Traffic Engineering with
Traditional IP Routing Protocols

           Manoj Ganesan
Aim of the Paper
   Provides an overview of working with the
    traditional IP routing protocols.

   No modification to the routing protocols or the
    routers themselves.

   How to adapt link weights, based on a network-
    wide view of the traffic and topology within a

   Summarizing the results of techniques for
    optimizing OSPF/IS-IS weights to the prevailing

   In some sense, IP networks manage themselves.
     Adjusting sending rate depending on
     Routers compute new paths

   However these mechanisms do ensure efficiency.
       Eg: Under-utilized links.

   The focus of this paper would be traffic within a
    single AS (company, ISP, etc).
Intradomain Traffic Engineering

   Path selection based on Static Link Weights.

   Limitations of static link weights, at the outset –
     Limited routing scenarios.
     No adaption of link weights, basically.

   Expensive extensions have been proposed.

   Can modify static link weights to do the job
A simple example.

   Initial configuration: 3 units of load on (u,t)

   Local change to the weight of the congested link,
    increased to 2. => 2.5 units of load on (w,t).

   Global optimization of the link weights. Most
    optimal solution.
Advantages of using traditional OSPF
         Process of arriving at a good set of weights is
          handled externally from the routers.

         Modification of link weights is performed on a
          relatively coarse time scale.

         Centralized approach for setting routing
          parameters. Has the following advantages:
             Protocol stability.
             Low protocol overhead.

         Use link weights to express routing config:
             Compatibility with traditional shortest path IGPs.
             Concise representation.
Traffic Engineering Framework.

   Measure
       Should have a timely and accurate view of the current
        state of the network.
       Estimate of the volume of traffic between each pair of
   Model
   Control
       Appropriate commands to affected routers.
       Router updates its link-state database & floods the new
        value of the rest of the network.
       Each router computes new shortest paths.
       No frequent changes to link weights.
Performance properties

   Quantifying how well we can engineer traffic flow
    using traditional OSPF/IS-IS routing protocols.

   Link Utilization.

   Comparing solutions with OPT routing, and simple
    configurations like InvCapOSPF and UnitOSPF.

   Returning back to our original example.
       UnitOSPF and InvCapOSPF, utilization = 150% (u,v).
       Last diagram, utilization for u,v = 100%.
Performance comparison with a network –
wide objective
          Advanced OSPF, comes closest to OPT routing
           (only 3% worse utilization than OPT)

          Minimizing max link utilization might be too
           specific and localized.
              Unavoidable congested links.
              No penalty to solutions that force traffic to traverse very
               long paths.

          Advances OSPF has an additional objective.
              The cost of using a link increases with the utilization,
               with an explosive growth as the utilization exceeds 100%
              Network wide cost of a routing solution is then the sum
               of all link costs.
Link cost as a function of the load for a link capacity 1
Network-wide cost vs. demand for a proposed AT&T backbone
Max link utilization vs. demand with same weights as previous graph
Changing traffic demands
   Optimizing the weights for a single topology and
    traffic matrix is not sufficient.

   Robustness was tested by changing traffic
    matrices, fluctuations, errors, etc.

   Previous weight settings (for the original TM)
    continued to perform well.

   Optimizing weights across traffic matrices!
Few changes to Link Weights

   Changes to link weights are necessary in response
    to large shifts in traffic and certain router or link

   Fortunately, changing even a single link weight is
    often effective.
       For an operational AT&T topology, increasing a single
        weight from 1024 to 1025 could reduce max utilization
        by 8%.
       Also, existing IGP weights continued to do well even
        after a single link failure.

   An overview of how to modify existing IP protocols
    to work efficiently in case of traffic fluctuations.

   This approach treats traffic engineering as a
    networks operation task, rather than a
    responsibility of the underlying routing protocol.

   As mentioned before, modifying traditional
    protocols have many advantages over proposed
    traffic engineering extensions to these protocols.

To top