perfSONAR User Q&A and Other Discussion

[Andrew Lake <>]

Hi Casey,

perfSONAR supports having dual-stacked, single-names hosts and to graph the two things separately. The MeshConfig options to do this go under the test_spec and are “ipv6_only 1” or “ipv4_only 1” respectively. If you just setup tests without specifying one of these options AND you are using only hostnames you are at the mercy of DNS. BWCTL will prefer v6 but fallback to v4 if it can’t find a v6 address for both endpoints. In contrast,  if you specify the ipv6_only option, it will not do this fallback. If you are seeing sporadic v4 results, I would suspect that there is some type of DNS resolution issue with one or more of the endpoints. Being explicit about what you want the test to do should at least prevent your performance results from being tainted by this occurrence.

Also, if you are using the meshconfig to also generate a dashboard, the resulting graphs and dashboards will see ipv6_only and/or ipv4_only options and generate dashboards and graph URLs that honor those options (FWIW there is an ipversion GET parameter it sets to 4 or 6 in the graph URLs). Hopefully that clears up any confusion.

Thanks,

Andy

On March 29, 2016 at 1:16:13 PM, Casey Russell () wrote:

Thank you Sowmya,

     My hosts are configured via the mesh_config option so I'll search and see if I can set that (v6 only) option in the .conf file so that the build_json script will regenerate my mesh with IPv6 tests only (since I only have v6 functionality today anyway).  However, that only addresses the cosmetic part of my problem.

     It seems that I may have to give up my dream of PS being able to test dual-stacked, single-named hosts and graph the two protocols separately. 

     I guess I'm going to have to consider creating new DNS names for the v4 addresses and create a new mesh based on those.  It seems such a "early 2000's" way of doing things with a dual-stacked host.  Does anyone have a quick guide for changing the hostname of a PS test node?  how many places do I need to go in and change the hostname in the PS config files?  Any gotchas to be aware of?

Casey Russell

Network Engineer

Kansas Research and Education Network

2029 Becker Drive, Suite 282

Lawrence, KS  66047

(785)856-9820  ext 9809

On Fri, Mar 25, 2016 at 2:32 PM, Sowmya Balasubramanian <> wrote:

Hi Casey,

If you want to run ipv6 tests, you can do so by selecting "ipv6" option while configuring the tests through the GUI. This will force the tool to run using ipv6 address only.

And, as far as the old test results are concerned, they will will stop showing up in the GUI after 30 days or so.

The issue that you have mentioned in your email( https://code.google.com/p/perfsonar-ps/issues/detail?id=1064) is different from your problem. But the perfSONAR development team is planning to add more support for displaying and differentiating between ipv6 and ipv4 tests in the same graph. The issue is currently being tracked here: http://github.com/perfsonar/graphs/issues/3

Hope this helps!

Thanks,

Sowmya

On Fri, Mar 25, 2016 at 9:07 AM, Casey Russell <> wrote:

Group,

     A little over a year ago, I asked about a bug in what was probably 3.4.1? at the time.  Where I have testing hosts that are dual stack.  The single host name has a A record and a AAAA record in DNS.  This results (over time) in lots of junk entries in my testing results table.  The majority of the time, the tests run correctly, and collect data, store data and keep the data in the databases, indexed by the IPv6 address since that gets preference. 

     However, once in a while for one reason or another when a test is kicked off, the DNS lookup results in an A record returned instead of a AAAA.  That creates a one-time orphaned test result that creates a new test results line on my toolkit page.  It looks just like the actual test results for those two hosts except that it has IPv4 addresses listed instead of IPv6. 

     Micheal Johnson responded last year and said it was a known bug https://code.google.com/p/perfsonar-ps/issues/detail?id=1064  but I haven't seen anything new on it.

     Does anyone else have this problem?  Do you have instructions for manually removing these entries from the databases or cleaning up the mess it makes?  Any news on when this bug's going to get into the pipeline for fixing?  Just looking for an update, but more importantly (if it's going to be a while) a way to fix the cosmetic damage myself so I can quit explaining those lines to my users.

Casey Russell

Network Engineer

Kansas Research and Education Network

2029 Becker Drive, Suite 282

Lawrence, KS  66047

(785)856-9820  ext 9809