perfSONAR User Q&A and Other Discussion

[Jason Zurawski <>]

Hi Brian;

>> Oh, and minor bug report:
>> The link at http://software.internet2.edu/owamp/download.html whose text 
>> points to the owamp download, actually points to the /ndt/ directory as 
>> its target.

I will pass this on to the Internet2 team that maintains that site.  Thank 
you for pointing that out.  

I will attempt to answer your 3 unanswered questions:

> Q1: is it possible/reasonable/recommended to set up one central management 
> node running perfsonar toolkit, and then install just bwctl-server and 
> iperf and/or owamp on the other nodes?
> 
> (I am even thinking of things like remote OSX workstations)
> 
> I am testing toolkit 3.4.1 inside a VM, with two other hosts just running 
> bwctl-server and iperf, and it appears to work; but I wonder if there are 
> pitfalls I should be aware of.

This mode of operation will work fine.  From the standpoint of the regular 
testing infrastructure, it just needs a willing participant to bounce the 
test off of - it doesn’t matter if its a full toolkit or just daemons.  Using 
the ‘full’ toolkit normally comes down to a choice of how people want to 
manage the infrastructure.  Some sites like having the machines be uniform in 
software setup and choose to make the central host and the beacons all the 
same, others may do what you are attempting and just go with minimal software 
requirements.  

> Q2: is it possible to get perfsonar to schedule tests between pairs of 
> remote hosts?
> 
> bwctl allows third-party testing, i.e. you can specify a remote client and 
> a remote server.
> 
> In the GUI under "Configure Tests" you can select the local interface and 
> add "test members" hosts. It's not entirely clear, but I presume the tests 
> all run between your host and the members, not between the members (which 
> would be an N^2 explosion). Is that correct?
> 
> In my test with two other hosts, the results page shows only
> 
> src = <remote host>  dst = <perfsonar host>
> 
> i.e. no graphs between <remote1> and <remote2>
> 
> In the FAQ, I see a mention in passing of a "mesh configuration", but 
> googling for that term turns up a google-code project which seems to be 
> between a mesh of perfsonar hosts.


“It depends”:

 - OWAMP will never support 3rd party operation on it’s own, unfortunately, 
due to the structure of the protocol.  BWCTL does support this mode for the 
tools it calls (e.g. traceroute, iperf, and the owamp client - owping).  

 - The GUI itself on the toolkit wouldn’t be able to schedule what you want 
(your guess was correct - the GUI assumes one end is the current machine)

 - The ‘mesh’ configuration also wouldn’t work for a 3rd party mode (also 
assumes one end is the current host).  

The one way that this could work is if you are comfortable with hand hacking 
one of the internal files.  The file in question is 
“/opt/perfsonar_ps/regular_testing/etc/regular_testing.conf”.  You may want 
to set up some regular tests with the GUI (if you haven’t done so already) 
and see how that file starts to become structured.  It should then be 
possible to hand edit that to substitute in remote hosts for the ‘self’ host. 
 It hasn’t been extensively tested (according to the developer I asked), so 
it may need some fiddling and require you to submit us bug reports (use the 
google code site - https://code.google.com/p/perfsonar-ps/issues/list).  

> Q6: Final supplementary question.
> 
> Is it possible to prevent a perfsonar instance registering itself in the 
> global 'directory of services'? Is not joining any communities enough to 
> stop this happening?

I suppose the easiest way would be to disable the service that registers:

        1) Edit 
'/opt/perfsonar_ps/ls_registration_daemon/etc/ls_registration_daemon.conf’ 
and add “disabled      1” to the top
        2) Restart the service: “sudo /etc/init.d/ls_registration_daemon 
restart”

Note that the service will have to stay ‘running’ (cron will make sure it 
gets started if you just stopped it), so this should prevent registrations 
(we think).  

> Also, If you have a private collection of perfsonar instances, can you run 
> your own lookup service and have them join that instead?


There is nothing that would stop you from doing this, this is not a supported 
mode of operation for us however, so we can’t really offer many documents on 
how to do it.  If I had to guess:

        1) install the sLS software somewhere 
(https://code.google.com/p/simple-lookup-service/)
        2) Hand edit your instances to point their ‘ls_instance’ variables to 
the one you setup (and set up a private bootstrap file, etc.)

Hope this helps;

-jason

> On Nov 26, 2014, at 4:21 PM, Brian Candler 
> <>
>  wrote:
> 
> On 26/11/2014 13:03, Brian Candler wrote:
>> Q3: In the main results page, I see a throughput figure src-dst, but 
>> dst-src is "n/a". Is it only measuring incoming traffic then? Is it 
>> possible to get measurement in the opposite direction as well?
>> 
>> I don't see how to configure this when I edit the configuration of a 
>> throughput test.
>> 
>> Q4: Probably a silly question, but when I view a throughput graph, I see a 
>> large graph at the top and a small graph at the bottom (with no vertical 
>> axis legend). What's the bottom graph showing?
>> 
> I found the description of the bottom time selector at 
> http://docs.perfsonar.net/using_graphs.html, so that solves Q4.
>> Q5: I have scheduled ping tests, but I can't see the results anywhere. The 
>> main page shows n/a for packet loss and latency. Is there anything which 
>> the remote hosts need for this to work? 
> I can answer Q3 and Q5 myself now.
> 
> My two other test endpoints are Ubuntu 14.04 (Trusty). The version of bwctl 
> in the those repositories is old (1.4.1), and either doesn't have or wasn't 
> compiled with bwping and bwtraceroute support. It could only drive iperf 
> (not iperf3) and nuttcp.
> 
> So I built iperf3 and bwctl from the 15.04 (Vivid) source packages. After 
> deploying these, I now get graphs for "reverse throughput" and "ping", in 
> addition to "throughput" that I had before. That solves Q3 and Q5 - 
> although I still don't get "reverse ping". But installing owampd (which I 
> compiled from source) I am able to get the loss and latency graphs.
> 
> So it looks in principle like I could use existing stations as endpoints. 
> However I get a warning that "STA_NANO should be set": does this mean that 
> the kernel and/or ntpd supplied as part of the perfsonar toolkit are 
> specially optimised?
> 
> Oh, and minor bug report:
> The link at http://software.internet2.edu/owamp/download.html whose text 
> points to the owamp download, actually points to the /ndt/ directory as its 
> target.
> 
> Still interested in answers to Q1, Q2 and Q6 :-)
> 
> Regards,
> 
> Brian.