/p4paxos-demo

Vagrant VM containing P4Paxos source code and its dependencies

Primary LanguagePython

Setup a Virtual Machine

You must install vagrant before running the following commands:

cd p4paxos-demo
vagrant up

##Connect to the VM From the p4paxos-demo directory, type:

vagrant ssh

Demo

After connecting to the VM, change to the p4paxos/bmv2 directory, and start the demo:

cd p4paxos/bmv2
sudo ./run_demo.sh

From the mininet prompt, type the following command to open a terminal on host h4:

mininet> xterm h4

Testing

In the h4 terminal, execute scripts/test.sh, which is reproduced below:

#!/bin/bash
curl -X POST http://10.0.0.1:8080/put -d key="ciao" -d value="hello"
curl -X GET  http://10.0.0.1:8080/get?key="ciao"
curl -X GET  http://10.0.0.1:8080/get?key=nonexist

You should see the following output:

Success
hello
None

Success means that the request to put the pair (ciao, hello) in the KeyValue store has succeeded. Similarly, hello is the value returned by the request to get the value for the key ciao. And None is the result of the request that gets a nonexistent key.

Or, start Firefox web browser in h4 terminal

firefox &

In the firefox browser, visit 10.0.0.1:8080. Then, you can try to Get or Put a (key, value) pair.

Simulating Failures

In this demo, there are two replicas (learners) running on h2 and h3. The service is still alive if a minority of the switches or replicas crash.

We can simulate the link failure by running the link command in Mininet:

mininet> link h2 s2 down

Or, simulate a server failure by stopping the server process:

mininet> h3 kill %python

If you perform another test similar to the one in section Testing, you should observe the same output, because node h2 still receives a majority of accepted requests from s3 and s4 switches. If the link from h2 to s3 is also down. Then the service is disrupted because the node h2 can't decide which request is selected for the current instance.

In mininet, type:

mininet> link h2 s3 down

and in h4's terminal, type:

curl -X GET  http://10.0.0.1:8080/get?key="ciao"

The request should fail.

Configuration

In the bmv2/scripts/paxos.cfg configuration file:

[instance]
count=65536

[timeout]
second=0

  • The count variable in the instance section configures the maximum number of requests that learners will handle.

  • The second variable in the timeout section configures the number of seconds to run the learners. If timeout is set to 0, the learners do not terminate.