perfSONAR User Q&A and Other Discussion

[Brian Candler <>]

On 06/02/2015 18:04, Lawrence Hitchens wrote:
> However, after reading some of the documentation, I think it is 
> over-complicated and effectively killing ants with a sledgehammer. The 
> multiple components (CA, MP, MA...etc) are vaguely documented and 
> there are multiple ways to deploy them. 
Like all good modular software, there are components which you can use 
individually or together.

The low-level tools for testing performance are:
* iperf3 for throughput (or you can use the older iperf, or nuttcp)
* owamp/owampd for one-way latency and packet loss measurements

You can run these directly at the command line. For example, to do an 
iperf3 test then at one side you run
    iperf3 -s
and at the other side you run
    iperf3 -c <ip-of-other-side>

But it means you need to login to both nodes to do this.

You could of course leave iperf3 -s running in daemon mode at one end if 
you wish, but the problem is that anyone can make a test to it at any 
time; if two people are testing simultaneously then they will be 
fighting with each other for bandwidth, and the results become meaningless.

So the next layer is bwctl. bwctl solves the following problems:

- you don't need to login to both sides to run a test; indeed, you don't 
have to login to either. If you are running bwctld on node A and bwctld 
on node B, you can trigger a test from node A to node B by running bwctl 
on node C.

- tests cause reservation time windows, so two tests can't run 
concurrently, no matter who is requesting them

- you can apply policies (bwctl limits) as to who can or cannot run tests

Note that you can install bwctl, owampd and iperf3 as standalone packages

That's great, however bwctl by itself will only run a test when you ask 
it to.

The next layer is the full perfsonar_ps toolkit, which gives you:
1. regular_testing. This allows you to configure classes of tests, with 
lists of endpoints, and have them run automatically at predefined intervals.
2. data storage (esmond), for all the test results to be archived
3. visualisation, i.e. draw graphs from your esmond data to show to your 
manager.

You can run tests between a perfsonar_ps toolkit node and a remote node 
which is only running bwctl and owampd.

I hope that makes the architecture clearer.

Regards,

Brian Candler.