I have two virtual machines connected within the same LAN,
I run my master with :
./ptpd2 -W -b eth0
and my slave with :
./ptpd2 -g -b eth0 -f output
Can you help me to understand the output: the clock drift stucks very fast to -512 us and stays here for a while… Then it increases:
2012-07-03 14:17:07.398245, slv, 080027fffe17cccd(unknown)/01, 0.000583198, 0.002149048, -0.001771374, 0.002898588, -105026, S
2012-07-03 14:17:07.567649, slv, 080027fffe17cccd(unknown)/01, 0.000581117, 0.002149048, -0.001892621, 0.002898588, -105026, D
2012-07-03 14:17:08.398672, slv, 080027fffe17cccd(unknown)/01, 0.000581117, 0.002490955, -0.001892621, 0.003247637, -102536, S
2012-07-03 14:17:08.880006, slv, 080027fffe17cccd(unknown)/01, 0.000582734, 0.002490955, -0.001514965, 0.003247637, -102536, D
2012-07-03 14:17:09.397878, slv, 080027fffe17cccd(unknown)/01, 0.000582734, 0.002500942, -0.001514965, 0.002918099, -100036, S
Then the clock drift increases again until ~140 us, and decreases again until ~80 us (then it is more or less stable).
Why such variations ?
What do you think about the results, is this a normal behavior ?
Why is the drift unable to reach a zero value ? (I never get a lstn_reset even when it's stable)
I can put a graph of what it looks like if it's not clear : )
Thank you !
Emilien
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
That's correct. Clocks inside of VMs are notoriously inaccurate and there really is no good way to use something like NTP or PTP within them, other than, perhaps, as a client.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2012-09-05
Hi, I modified de ptpd original source code so I can run it on a different plateform/different OS which has hardware timestamping. This plateform is running as a Master and I'm running a slave on a regular Linux. I still doesn't understand really well the output. I ran the R script provided in the "tools" directory and got these figures:
What do you think? Can this help to conclude on the accuracy of my slave's clock? If not, then how do I measure this? If you need the raw output, I can provide it.
Best regards,
Emilien
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I have two virtual machines connected within the same LAN,
I run my master with :
./ptpd2 -W -b eth0
and my slave with :
./ptpd2 -g -b eth0 -f output
Can you help me to understand the output: the clock drift stucks very fast to -512 us and stays here for a while… Then it increases:
2012-07-03 14:17:07.398245, slv, 080027fffe17cccd(unknown)/01, 0.000583198, 0.002149048, -0.001771374, 0.002898588, -105026, S
2012-07-03 14:17:07.567649, slv, 080027fffe17cccd(unknown)/01, 0.000581117, 0.002149048, -0.001892621, 0.002898588, -105026, D
2012-07-03 14:17:08.398672, slv, 080027fffe17cccd(unknown)/01, 0.000581117, 0.002490955, -0.001892621, 0.003247637, -102536, S
2012-07-03 14:17:08.880006, slv, 080027fffe17cccd(unknown)/01, 0.000582734, 0.002490955, -0.001514965, 0.003247637, -102536, D
2012-07-03 14:17:09.397878, slv, 080027fffe17cccd(unknown)/01, 0.000582734, 0.002500942, -0.001514965, 0.002918099, -100036, S
Then the clock drift increases again until ~140 us, and decreases again until ~80 us (then it is more or less stable).
Why such variations ?
What do you think about the results, is this a normal behavior ?
Why is the drift unable to reach a zero value ? (I never get a lstn_reset even when it's stable)
I can put a graph of what it looks like if it's not clear : )
Thank you !
Well… The more I try the more I have weird results :s
But I guess trying it inside a virtual machine is irrelevant, as it is for NTP:
http://support.ntp.org/bin/view/Support/KnownOsIssues#Section_9.2.2.
Hi,
It works like a charm with two non-virtualized linux on the same LAN
: )
That's correct. Clocks inside of VMs are notoriously inaccurate and there really is no good way to use something like NTP or PTP within them, other than, perhaps, as a client.
Hi, I modified de ptpd original source code so I can run it on a different plateform/different OS which has hardware timestamping. This plateform is running as a Master and I'm running a slave on a regular Linux. I still doesn't understand really well the output. I ran the R script provided in the "tools" directory and got these figures:
$ Rscript -slave stats.R myFile
Offset
min: -7.1123e-05 max: 7.6198e-05 median: 1.25e-07 mean: -2.740807e-07 std dev: 1.334304e-05 variance: 1.780367e-10
Delay
min: 7.7412e-05 max: 9.0117e-05 median: 8.467e-05 mean: 8.451698e-05 std dev: 2.518402e-06 variance: 6.342351e-12
Master -> Slave
min: 2.909e-06 max: 0.000197358 median: 8.547e-05 mean: 8.438647e-05 std dev: 1.851839e-05 variance: 3.429306e-10
Slave -> Master
min: 4.4949e-05 max: 0.000135107 median: 8.467e-05 mean: 8.475056e-05 std dev: 8.112882e-06 variance: 6.581885e-11
What do you think? Can this help to conclude on the accuracy of my slave's clock? If not, then how do I measure this? If you need the raw output, I can provide it.
Best regards,