perfsonar-user - Re: [perfsonar-user] XML for ignoring the first few seconds of Loss tests
Subject: perfSONAR User Q&A and Other Discussion
List archive
- From: Andrew Lake <>
- To: John Emery <>,
- Subject: Re: [perfsonar-user] XML for ignoring the first few seconds of Loss tests
- Date: Wed, 6 Mar 2019 21:43:27 +0100
Hi,
Your loss graphs are generated by OWAMP that uses UDP so there is no loss as a result of the TCP ramp-up to which I think you are referring. Spikes of 600 packets mean 100% packet loss for the 1 minute sample (600 packets at 10 packets per second = 60 seconds per your config).
What that likely indicates is that the full UDP port range of 8760-9960 is not open on one end of the test. Likely some or most of those ports are open so the test works fine, but occasionally your test picks a port from that range that is blocked so none of the test packets make it through. You can test it by trial and error using the owing command’s -P option to give it a smaller range of ports and see what fails. Example that strings it to 9900-9960:
owping -P 9900-9960 host.example
It could of course also be a legitimate packet loss event, but if you are seeing 600 packets exactly, I would start with the firewall given that the 600 packets aligns with the boundary of a test session and is the point at which it would switch to a new port.
Thanks,
Andy
On March 6, 2019 at 3:27:40 PM, John Emery () wrote:
We have a great Perfsonar/Maddash Map setup with a central management pSconfig
json file. I have noticed that we are getting frequent small spikes in the
Loss graphs and when I dig into the details they almost always lose about
600ish packets. This leads me to believe that this is standard "ramp up" loss
that is expected. I would like to eliminate the ramp up period of the test so
my graphs reflect smooth operation not spikey which is a little hard to
explain away to non technical folks. I cannot find anywhere in the
documentation the correct XML to insert into my probe.json file to ignore the
first few seconds of a test. If I configure a probe manually through the
probe GUI ignoring the first few seconds of a test is an option. Below are my
test tasks.. Any suggestions would be appreciated.
"tests": {
"throughput_test": {
"type": "throughput",
"spec": {
"source": "{% address[0] %}",
"dest": "{% address[1] %}",
"duration": "PT30S"
}
},
"latencybg_test": {
"type": "latencybg",
"spec": {
"source": "{% address[0] %}",
"dest": "{% address[1] %}",
"packet-interval": 0.1,
"packet-count": 600
}
},
"trace_test": {
"type": "trace",
"spec": {
"source": "{% address[0] %}",
"dest": "{% address[1] %}"
}
},
"rtt_test": {
"type": "rtt",
"spec": {
"source": "{% address[0] %}",
"dest": "{% address[1] %}",
"count": 10,
"interval": "PT1S",
"length": 1000
}
}
},
Best regards,
John
--
To unsubscribe from this list: https://lists.internet2.edu/sympa/signoff/perfsonar-user
- [perfsonar-user] XML for ignoring the first few seconds of Loss tests, John Emery, 03/06/2019
- Re: [perfsonar-user] XML for ignoring the first few seconds of Loss tests, Andrew Lake, 03/06/2019
Archive powered by MHonArc 2.6.19.