perfSONAR User Q&A and Other Discussion

["Andrew Lake" <>]

Hi Brian,

Do you have “send_only 1" set for you ping tests in regular_testing.conf? All ping tests will use bwping, but if your local host is the one sending the ping it will run an endpointless test (the bwping -E option) which means that it won’t try to talk to BWCTL on the other side, it will just fire off the ping. It sound slike it is trying to do a revese test which will happen if you have receive_only set or if neither is set (in which case it does bidirectional I believe).

Thanks,

Andy

On Thu, Nov 5, 2015 at 11:44 AM, Brian Candler <> wrote:

On 05/11/2015 16:23, Brian Candler wrote:
> bwctl: exec_line: ping6 -c 10 -i 1.000000 -s 1000 -m 255 -S
> 2001:db8:1:64::252 2001:db8:0:2::254
OK, I can replicate this at the command line. On the perfsonar 3.5 node:

$ bwping -t 255 -c <remote-OSX-node-running-bwctl> # works fine, but:

$ bwping -t 255 -s <remote-OSX-node-running-bwctl>
bwping: Using tool: ping
bwping: 21 seconds until test results available

SENDER START
ping6: setsockopt(IPV6_RECVPATHMTU): Operation not permitted
bwctl: local tool exited before expected with status=1

SENDER END


This seems to be another symptom of perfsonar changing the tests around,
as I already saw for traceroute.

(1) What I had *configured* was a ping test on the perfsonar node, with
the target as the OSX node.

(2) What I *expected* was simply for the perfsonar node to ping the OSX
node. This was working in perfsonar 3.4.x

(3) What appears to be *happening now* is that perfsonar is using bwping
to ask bwctld on the remote host to run a ping test *from* the remote
node *back* to the perfsonar node.

This is problematic on several levels:

- I didn't want bwping to be used. "ping" tests in perfsonar are not
very useful if they cannot be used to ping dumb hosts that are not
running bwctl

In fact, the perfsonar 3.4 node I have is being used to ping some dumb
routers; this will prevent me from upgrading that box to 3.5.

- The pings are now being originated from the OSX side; I expected the
pings to originate from the perfsonar side

- OSX bwctld for ping is broken. As it turns out, "-m 255" is the
correct way to specify TTL 255 on OSX, but there is some sort of socket
permissions problem which prevents it working.

These problems would go away if perfsonar went back to its original
approach of originating traceroute and ping from the perfsonar node, not
from the remote node.


This still leaves the issue of why IPv4 pings aren't working, and I
haven't got to the bottom of that yet.

Regards,

Brian.