Differences between revisions 8 and 14 (spanning 6 versions)

BattleMeshV5 Tests

Contents

BattleMeshV5 Tests
1. General Rules
2. Tests

General Rules

same routing protocol revision and configuration for every test
routing protocols are tested in turn (i.e. not altogether)
cpu and memory consumption should be tracked during each test
no points, just results and a final analysis

Tests

Be Quick or Be Stable Test

To be performed on a subset of the nodes.

 +----+
 | L1 |-----(A) : : : : : : (B)
 +----+      :       r1      |
             :               |
             :               |
             : r2            |
             :               |
             :               |       +----+
            (C)-------------(D)------| L2 |
                                     +----+

L1, L2: laptops
A, B, C: wireless nodes
link A-B uses association rate r1
link A-C uses association rate r2
r2 > r1
- and | are cables
: are wireless links

Steps

5x for each protocol

perform a continuous traceroute (mtr or whatever) from L1 to L2
load the network with UDP (iperf or whatever) from L1 to L2
start a 11 minutes download from L2 to L1
see how many times (if any) the route flaps in 11 minutes
see if the routing protocol prefers the slow or the fast route

Goals

Check the stability of the routing protocol, and if the metric is effective.

References

http://www.youtube.com/watch?v=fZLv2G0Hhn4

The Channel Surfer Test

To be performed on a subset of the nodes.

 +----+
 | L1 |-----(A)
 +----+      :
           x :
             :
            (B) : : : : : : (D)
             |       x       |
             |               |
             |               |      +----+
            (C) : : : : : : (E)-----| L2 |
                     y              +----+

L1, L2: laptops
A, B, D: nodes set on channel x
C, E: nodes set on channel y
- and | are cables
: are wireless links

Steps

only 1x for each protocol

perform a continuous traceroute (mtr or whatever) from L1 to L2
load the network with TCP (iperf) from L1 to L2 for 21 minutes
sleep until the end of the iperf stream
see for which fraction of the time the routing protocol chooses the best route (L1->A->B->C->E->L2)
see the amount of transferred bytes

Goal

See if the routing protocol prefers the channel-changing route, by design or by just because is the best path.

Convergence Time Test

To be performed on the whole mesh.

           .~.~.~.~.~.~.~.~.~.~.~.       
 +----+   (                       ) : (GW1)-+
 | L1 | :(                         )        |   
 +----+   (                       )         |   +----+
         (                         )        +---+ L2 |
          (                       )         |   +----+
         (                         )        |
          (                       ) : (GW2)-+
         (        MESH CLOUD       )     
          .~.~.~.~.~.~.~.~.~.~.~.~.

GW1 and GW2: gateways announcing the address of L2
The metric from L1 to GW1 is significantly better than the metric from L1 to GW2

Steps

repeat 3x for each protocol

traceroute from L1 to L2 and verify that the path goes through GW1
start pinging from L1 to L2
start a continuous traceroute (mtr or whatever) from L1 to L2
turn off GW1
count the number of pings from the first lost ping to the first subsequent successful ping
after 2 minutes from its turning off, turn back on GW1 and start a stopwatch
measure the time needed to the network to use GW1 again

Goal

Just convergence time.

Don't Cross the Streams Test

To be performed on the whole mesh

 +----+            .~.~.~.~.~.~.~.~.~.~.~.            +----+
 | L1 |-----(A) : (                       ) : (C)-----| L3 |
 +----+          (                         )          +----+
                  (                       )
                 (          MESH           )
                  (         CLOUD         )
 +----+          (                         )          +----+
 | L4 |-----(D) : (                       ) : (B)-----| L2 |
 +----+          (                         )          +----+
                  .~.~.~.~.~.~.~.~.~.~.~.~.

L1, L2, L3, L4: laptops
A, B, C, D: edge wireless nodes

Steps

only 1x for each protocol

start, at the same time:
1. a 21 minutes TCP stream between L1 and L2
2. a 21 minutes TCP stream between L3 and L4
sleep for 21 minutes
measure the total number of bytes transferred

Goal

See if the routing protocol can choose the paths for the two streams that maximizes throughput.

References

http://www.youtube.com/watch?v=jyaLZHiJJnE

-  ⇤ ← Revision 8 as of 2012-03-01 12:37:16 → 
  Size: 4958
  Editor: Clauz
  Comment:
+   ← Revision 14 as of 2012-03-31 12:34:23 → ⇥
  Size: 4959
  Editor: Clauz
  Comment:
-Deletions are marked like this.
+Additions are marked like this.
 Line 26:
-             :            +----+
            (C)-----------| L2 |
                          +----+
+             :               |       +----+
            (C)-------------(D)------| L2 |
                                     +----+
 Line 40:
-only 1x for each protocol
+x for each protocol
 Line 43:
-. start a 21 minutes download from L2 to L1
 1. see how many times (if any) the route flaps in 21 minutes
+. start a 11 minutes download from L2 to L1
 1. see how many times (if any) the route flaps in 11 minutes
 Line 62:
- | L1 |-----(A) : : : : : : (B)
 +----+      :       x       :
             :               :
           x :               : x
             :               :   
             :               :
             :               :
            (C)             (E)       
             |               |        +----+
            (D) : : : : : : (F)-------| L2 |
                     y                +----+
+ | L1 |-----(A)
 +----+      :
           x :
             :
            (B) : : : : : : (D)
             |       x       |
             |               |
             |               |      +----+
            (C) : : : : : : (E)-----| L2 |
                     y              +----+
-Line 76:
+Line 75:
- * A, B, C, E: nodes set on channel x
 * D, F: nodes set on channel y
+ * A, B, D: nodes set on channel x
 * C, E: nodes set on channel y
-Line 83:
+Line 82:
-. load the network with TCP (iperf or whatever) from L1 to L2
-Line 85:
+Line 83:
-. sleep for 21 minutes
 1. see for which fraction of the time the routing protocol chooses the best route (L1->A->C->D->F->L2)
+. load the network with TCP (iperf) from L1 to L2 for 21 minutes
 1. sleep until the end of the iperf stream
 1. see for which fraction of the time the routing protocol chooses the best route (L1->A->B->C->E->L2)
 1. see the amount of transferred bytes
 Line 98:
-    (A) : (                       ) 
         (                         )     
          (                       )
         (           (N)           )
          (                       )
         (                         )     
          (                       ) : (B)
+ +----+   (                       ) : (GW1)-+
 | L1 | :(                         )        |   
 +----+   (                       )         |   +----+
         (                         )        +---+ L2 |
          (                       )         |   +----+
         (                         )        |
          (                       ) : (GW2)-+
 Line 109:
-==== Steps - part 1 ====
repeat 3x for each protocol
 1. traceroute from one end (node ''A'') to the other (node ''B'') of the mesh until the path is stable
 1. find a critical node ''N'' in the path so that 
   a. has good quality links with its neighbors (possibly more than one)
   a. if shut down the network does not split
 1. start a ping from ''A'' to ''B''
 1. turn off ''N''
 1. count the number of pings from the first lost ping to the first subsequent successful ping
+ * ''GW1'' and ''GW2'': gateways announcing the address of ''L2'' 
 * The metric from L1 to GW1 is significantly better than the metric from L1 to GW2
-Line 120:
+Line 113:
-==== Steps - part 2 ====
 1. start a continuous traceroute (mtr or whatever) from node ''A'' to node ''B''
 1. turn back on node ''N'' and start a stopwatch
 1. measure the time needed to get the node ''N'' (back) into the traceroute
+==== Steps ====
repeat 3x for each protocol
 1. traceroute from ''L1'' to ''L2'' and verify that the path goes through ''GW1''
 1. start pinging from ''L1'' to ''L2''
 1. start a continuous traceroute (mtr or whatever) from ''L1'' to ''L2''
 1. turn off ''GW1''
 1. count the number of pings from the first lost ping to the first subsequent successful ping
 1. after 2 minutes from its turning off, turn back on ''GW1'' and start a stopwatch
 1. measure the time needed to the network to use ''GW1'' again
-Line 154:
+Line 152:
-   a. a TCP stream between L1 to L2
   a. a TCP stream between L3 to L4
   a. a stopwatch
+   a. a 21 minutes TCP stream between L1 and L2
   a. a 21 minutes TCP stream between L3 and L4