Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
MSTP-TE Load Balancing: Some results Benchmarking Carrier Ethernet Technologies Session AI.3 Krakow, Poland - April 30, 20...
Goals  &  Assumptions
Input data <ul><li>Work done in collaboration with ALF </li></ul><ul><li>Topology </li></ul><ul><ul><li>Define a common &q...
Experiments
Some definitions <ul><li>For each &quot;variation&quot;, different experiments were done </li></ul><ul><li>CL/CO </li></ul...
Variations of the topology for the experiment <ul><li>V0 </li></ul><ul><ul><li>Reference Topology, aggregation network ful...
Partial results
Results criteria and format <ul><li>Three major indicators chosen jointly with ALF  (1 curve, two values) </li></ul><ul><u...
V0 <ul><li>Analyze (to be discuss) </li></ul><ul><ul><li>Multicast give a non negligible gain against &quot;multi unicast&...
V1b <ul><li>Analyze </li></ul><ul><ul><li>No &quot;big&quot; difference between CO and CL </li></ul></ul><ul><ul><li>LB gi...
  Let us try to conclude …   ... keeping the door open for discussion
Some precautions to consider <ul><li>What is a good &quot;Load Balancing&quot; ??? </li></ul><ul><ul><li>Regarding the nee...
Preliminary conclusions <ul><li>With MSTP, paths are constrained to follow &quot;trees&quot;  </li></ul><ul><ul><li>At fir...
Thanks for your attention
Upcoming SlideShare
Loading in …5
×

Ethernet Bridging (MSTP, SPB/PLSB)

885 views

Published on

  • Be the first to comment

  • Be the first to like this

Ethernet Bridging (MSTP, SPB/PLSB)

  1. 1. MSTP-TE Load Balancing: Some results Benchmarking Carrier Ethernet Technologies Session AI.3 Krakow, Poland - April 30, 2008 Rémi Clavier - Orange Labs COBRA Project
  2. 2. Goals & Assumptions
  3. 3. Input data <ul><li>Work done in collaboration with ALF </li></ul><ul><li>Topology </li></ul><ul><ul><li>Define a common &quot;Reference Topology&quot; </li></ul></ul><ul><ul><li>Define &quot;variations&quot; from this topology </li></ul></ul><ul><ul><ul><li>To have a more exhaustive analysis </li></ul></ul></ul><ul><ul><ul><li>To take into account realistic FT aggregation topologies </li></ul></ul></ul><ul><li>Matrix </li></ul><ul><ul><li>Define a common (more or less) realistic traffic matrix including </li></ul></ul><ul><ul><ul><li>http,HSI, VoIP </li></ul></ul></ul><ul><ul><ul><li>Pear to Pear </li></ul></ul></ul><ul><ul><ul><li>IPTv </li></ul></ul></ul><ul><ul><ul><li>VPNs </li></ul></ul></ul><ul><ul><li>Define variation from this Matrix </li></ul></ul><ul><ul><ul><li>To try to catch P2P / P2MP influence </li></ul></ul></ul>
  4. 4. Experiments
  5. 5. Some definitions <ul><li>For each &quot;variation&quot;, different experiments were done </li></ul><ul><li>CL/CO </li></ul><ul><ul><li>CL (as Connection Less) </li></ul></ul><ul><ul><ul><li>The Control Plane is the 802.1Q one (MSTP with 1, 3 or 6 trees) </li></ul></ul></ul><ul><ul><ul><li>The Management plane set the MSTP's parameters of the Control Plane </li></ul></ul></ul><ul><ul><li>CO (as Connection Oriented) </li></ul></ul><ul><ul><ul><li>No Control Plane, the forwarding is positioned by the management plane </li></ul></ul></ul><ul><li>LB/SP (for CO only) </li></ul><ul><ul><li>LB (aka Load Balancing) </li></ul></ul><ul><ul><ul><li>The tool tries to optimize (maximize) the load balancing over the full network </li></ul></ul></ul><ul><ul><li>SP (aka Shortest Path) </li></ul></ul><ul><ul><ul><li>The tool tries to optimize (minimize) the sum of the hops for all flows over the full network </li></ul></ul></ul><ul><ul><ul><ul><li>A route is &quot;acceptable&quot; only if no link is overloaded over the full Network </li></ul></ul></ul></ul><ul><li>MU/UN </li></ul><ul><ul><li>MU (aka Multicast) </li></ul></ul><ul><ul><li>UN ( aka Unicast) </li></ul></ul>
  6. 6. Variations of the topology for the experiment <ul><li>V0 </li></ul><ul><ul><li>Reference Topology, aggregation network fully meshed </li></ul></ul><ul><ul><li>All links 10 G </li></ul></ul><ul><ul><li>TV dispatcher directly connected to the aggregation Network </li></ul></ul><ul><li>V0b </li></ul><ul><ul><li>Reference Topology , aggregation network fully meshed </li></ul></ul><ul><ul><li>All DSLAM links 1G ; All other links (except one) of the Aggregation Network at 10G </li></ul></ul><ul><ul><li>TV dispatcher directly connected to the aggregation Network </li></ul></ul><ul><li>V1 </li></ul><ul><ul><li>Reference Topology , aggregation network fully meshed </li></ul></ul><ul><ul><li>All links 10 G </li></ul></ul><ul><ul><li>TV dispatcher outside the aggregation Network (core network) </li></ul></ul><ul><li>V1b </li></ul><ul><ul><li>Reference Topology , aggregation network fully meshed </li></ul></ul><ul><ul><li>All DSLAM links 1G ; All other links (except one) of the Aggregation Network at 10G </li></ul></ul><ul><ul><li>TV dispatcher outside the aggregation Network (core network) </li></ul></ul><ul><li>V3 </li></ul><ul><ul><li>Reference Topology but Aggregation network not meshed (Ring Aggregation topology) </li></ul></ul><ul><ul><li>All DSLAM links 1G ; All other links (except one) of the Aggregation Network at 10G </li></ul></ul><ul><ul><li>TV dispatcher outside the aggregation Network (core network) </li></ul></ul>
  7. 7. Partial results
  8. 8. Results criteria and format <ul><li>Three major indicators chosen jointly with ALF (1 curve, two values) </li></ul><ul><ul><li>PFD curve </li></ul></ul><ul><ul><ul><li>Probability Density Function </li></ul></ul></ul><ul><ul><ul><ul><li>The probability that the load (in term of capacity of the link) is inside a given interval </li></ul></ul></ul></ul><ul><ul><ul><ul><li>The CDF is the integral of the PDF and not used directly to compare results </li></ul></ul></ul></ul><ul><ul><li>The ME value </li></ul></ul><ul><ul><ul><li>The average of the PDF function </li></ul></ul></ul><ul><ul><ul><ul><li>May give information about the fact that the less loaded links are preferentially chosen </li></ul></ul></ul></ul><ul><ul><li>The SD value </li></ul></ul><ul><ul><ul><li>The root mean square of the PDF curve </li></ul></ul></ul><ul><ul><ul><ul><li>shows the dispersion of the load of links around the full network </li></ul></ul></ul></ul>
  9. 9. V0 <ul><li>Analyze (to be discuss) </li></ul><ul><ul><li>Multicast give a non negligible gain against &quot;multi unicast&quot; </li></ul></ul><ul><ul><li>For MU, no difference between CO and CL </li></ul></ul><ul><ul><li>For UN, CO seems better than CL </li></ul></ul><ul><li>FT Remarks (from detailed results) </li></ul><ul><ul><li>3 trees are enough (no specific gain with 6 trees) </li></ul></ul><ul><ul><ul><li>Fully Meshed Aggregation network </li></ul></ul></ul><ul><ul><ul><li>All Links 10G </li></ul></ul></ul><ul><ul><ul><li>TV inside aggregation Network </li></ul></ul></ul>
  10. 10. V1b <ul><li>Analyze </li></ul><ul><ul><li>No &quot;big&quot; difference between CO and CL </li></ul></ul><ul><ul><li>LB give a better REm and a well better SD than SP in a CO context </li></ul></ul><ul><ul><li>PDF curve shows that CO/SP doesn't find a correct load balancing </li></ul></ul><ul><ul><li>Fully Meshed Aggregation network </li></ul></ul><ul><ul><li>All Links (except one) 10G for the aggregation Networks, DSLAM links 1G </li></ul></ul><ul><ul><li>TV dispatcher outside aggregation Network </li></ul></ul>
  11. 11.   Let us try to conclude … ... keeping the door open for discussion
  12. 12. Some precautions to consider <ul><li>What is a good &quot;Load Balancing&quot; ??? </li></ul><ul><ul><li>Regarding the needs of the operator </li></ul></ul><ul><ul><li>Regarding the technical context where analyzes are done </li></ul></ul><ul><ul><li>Regarding … everything else </li></ul></ul><ul><li>What does appear in the results? </li></ul><ul><ul><li>The intrinsic properties of the network's technologies? </li></ul></ul><ul><ul><li>… or the properties of tools used (optimizers)? </li></ul></ul><ul><li>Take care about results that are given </li></ul><ul><ul><li>for one kind of topology </li></ul></ul><ul><ul><li>For one traffic matrix </li></ul></ul><ul><ul><li>Although many variations are proposed </li></ul></ul>
  13. 13. Preliminary conclusions <ul><li>With MSTP, paths are constrained to follow &quot;trees&quot; </li></ul><ul><ul><li>At first sight, this constraint could decrease load balancing performances compared to ELS networks </li></ul></ul><ul><ul><li>But, with a TE tool, MSTP gives the same results as CO networks in terms of load balancing </li></ul></ul><ul><li>For a very loaded network and/or a network with links with different capacities of links </li></ul><ul><ul><li>A centralized optimization gives better results (load balancing) than optimization based solely on the calculation of a Shortest Path </li></ul></ul><ul><ul><li>The centralized optimization tool gives equivalent performances for CL or CO networks with LB routing (&quot;TE&quot;) algorithm </li></ul></ul><ul><li>Other &quot;well known&quot; properties of MSTP are not impacted by load balancing optimization </li></ul><ul><ul><li>Low cost </li></ul></ul><ul><ul><li>&quot;Bad&quot; convergence time </li></ul></ul><ul><ul><li>Compliant to standard </li></ul></ul><ul><ul><li>Natively multicast… </li></ul></ul>
  14. 14. Thanks for your attention

×