Improve perfSONAR support of IPv6

[laeti-tia]

Running a perfSONAR IPv6 only host has some limitations at the moment. perfSONAR support of IPv6 could be improved at least at 2 levels (but it's probably worth doing further investigations).

1. If a target has a AAAA record, pS should default to using IPv6 to do measurements, even if the --ip-version6 flag is not used in the pscheduler command
2. LS registration for IPv6 only hosts should be possible, the global LS host is only IPv4 at the moment.

(this was initially reported by Garret Peirce from UMaine through an indirect channel)

[laeti-tia]

For 1. we should review all tools and see which ones are RFC-6724 compliants and which ones are not. It would be interesting to add a page to our documentation describing the IPv6 support in perfSONAR.

For 2. we should add an IPv6 address to the global LS server, make sure all services are listening on it and publish it in an AAAA record and in our LS bootstrap information. Then make sure the ls-registration-deamon is using this new IPv6 address.

[laeti-tia]

Comment from Tim Chown:

In practice, software should be written mindful of RFC 6724 (one of mine!), so the sorting of addresses as returned by getaddrinfo() that return a list of addresses to the application will by default follow that. Or should. Then it’s up to the application to honour that ordering.

Happy eyeballs (RFC8305) explains how you can optimise performance (rather than consistently use one IP version) by trying both versions and using the first to complete/connect, typically giving IPv6 a small head start.

[mfeit-internet2]

Most of those problems fall to pScheduler, and there's a ticket open on it: perfsonar/pscheduler#855

[igarny]

Hi guys,

Please excuse me for the subjective classification.

Most relevant:
perfsonar/elmond#12
perfsonar/toolkit#448
perfsonar/archive#23

More loosely related:
perfsonar/psconfig#79
perfsonar/psconfig-web#217

[laeti-tia]

Also see perfsonar/toolkit#448 and perfsonar/docs#246