Clock discrepancy between IOSv nodes running concurrently

[d_b]

Hi,

I'm experiencing quite substantial clock discrepancy between IOSv nodes. IOL nodes behave much better in that regard, see results below from 4x IOSv + 4x IOL nodes running concurrently. Makes it quite challenging if not practically impossible to compare debug timestamps from multiple IOSv nodes.

Are you experiencing the same?
Is this a property of IOSv, emulation used in EVE, else?
Is there any way to fix it?

Thanks!


Clocks have been simultaneously synced with 'clock read', followed at a later time by 'show clock':

Four nodes running vios_l2-ADVENTERPRISEK9-M 15.2(4.0.55)E

*16:50:49.680 UTC Thu Oct 5 2017
*16:50:50.460 UTC Thu Oct 5 2017
*16:50:50.532 UTC Thu Oct 5 2017
*16:50:50.322 UTC Thu Oct 5 2017


Four nodes running I86BI_LINUXL2-ADVENTERPRISEK9-M 15.2(CML_NIGHTLY_20151103)FLO_DSGS7

*16:50:53.335 UTC Thu Oct 5 2017
*16:50:53.333 UTC Thu Oct 5 2017
*16:50:53.334 UTC Thu Oct 5 2017
*16:50:53.334 UTC Thu Oct 5 2017

[ramindia]

Hi,

I'm experiencing quite substantial clock discrepancy between IOSv nodes. IOL nodes behave much better in that regard, see results below from 4x IOSv + 4x IOL nodes running concurrently. Makes it quite challenging if not practically impossible to compare debug timestamps from multiple IOSv nodes.

Are you experiencing the same?
Is this a property of IOSv, emulation used in EVE, else?
Is there any way to fix it?

Thanks!


Clocks have been simultaneously synced with 'clock read', followed at a later time by 'show clock':

Four nodes running vios_l2-ADVENTERPRISEK9-M 15.2(4.0.55)E

*16:50:49.680 UTC Thu Oct 5 2017
*16:50:50.460 UTC Thu Oct 5 2017
*16:50:50.532 UTC Thu Oct 5 2017
*16:50:50.322 UTC Thu Oct 5 2017


Four nodes running I86BI_LINUXL2-ADVENTERPRISEK9-M 15.2(CML_NIGHTLY_20151103)FLO_DSGS7

*16:50:53.335 UTC Thu Oct 5 2017
*16:50:53.333 UTC Thu Oct 5 2017
*16:50:53.334 UTC Thu Oct 5 2017
*16:50:53.334 UTC Thu Oct 5 2017

You can set own NTP Server or use Public NTP Server to Sync.

R!

[d_b]

Did you succeed in solving IOSv clock deviation with syncing it with NTP? Please share your console output.

NTP doesn't kick free running software clock every single microsecond which would be needed to attain required precision so if the clock sucks in the first place then no external slow correction will make it precise.

My original question was about how come IOU/IOL gets clock ticks/reading from the underlying host infrastructure while IOSv is failing miserably to do the same thing.

[d_b]

Out of curiosity I checked with CSR 1000v nodes, their clocks don't run in step either.

Can anyone explain please how accurate time keeping on guest nodes is achieved with qemu in EVE?