PingDemo | Cougaar

NOTE: This file and all referenced files are included in the demo-ping.zip of the Cougaar.

This module contains the Cougaar “ping” demos.

There are six demos:

A basic “ping” test, where agent “A” sends a ping message to agent “B” once every 5 seconds.
A two-node “ping” test, which is identical to the above test, except the agents run in different JVMs.
A two-host “ping” test, which simply distributes the above JVMs (and their agents) across two hosts.
A single-node “ping” benchmark, where the the agents send ping messages as fast as possible. The “messages-per-second” throughput can be viewed by accessing a browser page.
A two-node “ping” profiled benchmark, where detailed performance metrics are logged every 10 seconds. Logged metrics include the throughput, inter-agent traffic matrix, and cpu load
A two-node “ping” where agent “A” sends to 11 other agents (i.e. a “fan-out” pattern)

For additional information, please see the Cougaar Overview.

configs/*.xml: The society files.
run/*.xml: The runtime files.
run/logging.props: The Log4j logging configuration file.
src/org/cougaar/demo/ping/*.java: The Java source code consists of:
- A “ping sender”
- A “ping receiver”
- A “ping servlet”

Installation:

Please follow the same installation steps as in the “Hello World” example, substituting “ping” for “hello”.

The “Hello World” README is available in the “cougaar-hello.zip”

Usage:

There are six demos:

A basic “ping” test:
1. Run:
```
cd $COUGAAR_RUNTIME_PATH
cougaar $COUGAAR_SOCIETY_PATH/configs/PingBasicSociety.xml PingBasicRuntime.xml
```
  This should print:
  [PingSender] – A: Sending ping 0 to B
  [DOTS] – +-
  [PingReceiver] – B: Responding to ping 0 from A
  [DOTS] – +-
  [PingSender] – A: Received response 0 from B
  [PingSender] – A: Will send ping 1 to B in 5 seconds
  [DOTS] – .
  …
2. Optionally view the servlet at
  http://localhost:8800/$A/ping
3. Press CTRL-C to stop the run.
A two-node “ping” test:
1. In one console window, run:
```
cd $COUGAAR_RUNTIME_PATH
cougaar $COUGAAR_SOCIETY_PATH/configs/PingTwoNodesSociety.xml PingTwoNodesRuntime.xml Node1
```
2. In a second console window, run:
```
cd $COUGAAR_RUNTIME_PATH
cougaar $COUGAAR_SOCIETY_PATH/configs/PingTwoNodesSociety.xml PingTwoNodesRuntime.xml Node2
```
3. The output should match the output of the single-node ping test. There may be a startup delay of up to a minute, due to the naming service.
4. Press CTRL-C to stop the run.
A two-host “ping” test:
1. Modify the $COUGAAR_SOCIETY_PATH/configs/PingTwoNodesRuntime.xml to change the host name from “localhost” to the host that will run “Node1″.
2. Perform the same steps as in the two-node test.If the agents are unable to communicate with one another, verify that both hosts can `/bin/ping` one another, that ports 8888 and 8800 are not blocked, and that `/bin/hostname -iv` prints the correct host names and IP addresses. If all else fails, please contact us.
A single-node “ping” benchmark:
1. Run:
```
cd $COUGAAR_RUNTIME_PATH
cougaar $COUGAAR_SOCIETY_PATH/configs/PingBenchmarkSociety.xml PingBenchmarkRuntime.xml
```
  Note that the above runtime XML file has been modified to disable the logging output.
2. Optionally view the servlet at
  http://localhost:8800/$A/ping
  Note the high “pings-per-second” throughput, due to the below “delayMillis=0″ setting.
3. Optionally view the metrics servlets, e.g.:
  http://localhost:8800/metrics/agent/load
  For a full listing of included metrics servlets, see: doc/OnlineManual/MetricsService/operation.html
4. Press CTRL-C to stop the run.
A two-node “ping” profiled benchmark:
1. In one console window, run:
```
cd $COUGAAR_RUNTIME_PATH
cougaar $COUGAAR_SOCIETY_PATH/configs/PingProfileSociety.xml PingProfileRuntime.xml Node1
```
2. In a second console window, run:
```
cd $COUGAAR_RUNTIME_PATH
cougaar $COUGAAR_SOCIETY_PATH/configs/PingProfileSociety.xml PingProfileRuntime.xml Node2
```
3. As in the single-node “ping” benchmark, the standard ping logging has been disabled, however, Node1 is configured to log detailed profiling output every 10 seconds.
  The logged metrics include various Cougaar-internal metrics:
  - inter-agent messaging traffic matrix
  - aggregate messaging throughput
  - blackboard sizes and number of add/change/remove operations
  - thread pool load
  - JVM heap size
  and system metrics:
  - cpu load (from /proc/loadavg)
  - memory load (from /proc/meminfo)
  - network stats (from /proc/net/dev)
  - overall system stats (from /usr/bin/vmstat)
4. Let both nodes run for a couple minutes, then examine the logs for the following interesting metrics:
  - “[tl_A]” is the number of messages sent by agent A to any target
  - “[tr_B]” is the number of messages received from agent B to any agent on Node1
  - “[tm_A__to__B]” is the number of messages sent between agents A and B
  - “[throughput]” is the number of message sent and received
  - “[bb_A__Relay]” is the number of relay add/change/removes
  - “[load_node_Node1]” is the cpu load average
  The “[throughput]” is a good overall performance metric. Here is example output:
```
[throughput] - #agent_send_count_per_second, total_send_count_per_second, agent_send_count, total_send_count,
[throughput] - 257.90, 258.10, 8190, 8205
```
  The above throughput rates should be twice the “/ping” servlet rate, since each ping iteration consists of a ping message and an “ack” message.
  
  For documentation on the other profiled metrics, please see the “profile” module code and javadoc.
5. Press CTRL-C in both console windows to stop the run.
A two-node “ping” fan:
1. Run the PingFanSociety.xml and PingFanRuntime.xml files, as noted in the above two-node tests. Examine the “/$A/ping” servlet to view the society progress.