I'm working on an NTP infrastructure project and trying to demonstrate the
benefits of bare metal (vs VMs running the same OS)for commodity lower-stratum
servers...but am having a hard time quantifying the difference.
For example, here are two nodes in the same subnet and with the same upstream
servers. The first is on reasonably recent server-class Dell hardware which is
otherwise idle. The second is running on a recent ESXi kernel with similar
hardware to the first. I have several stratum 3 hosts using both of these as
sources and there doesn't appear to be a clear winner in accuracy. If anything,
the VM tends to be chosen in source elections more often than not. Also, if you
look at the precision below, the bare metal is reporting 2^-22 and the VM is
2^-24. Am I missing something in my comparisons? Do I need to reduce the
variability of the upstream S1 sources to guarantee a more "level playing
field" and most importantly, what are the prime indicators of time signal
quality when comparing the difference between a hardware PC clock and a
virtualized one ?
--------------------------------------------------------------------
root@ntp02# ntpq -pn
remote refid st t when poll reach delay offset jitter
==============================================================================
*172.2*.8.33 .GPS. 1 u 21 64 77 14.851 0.018 1.290
+172.2*.0.33 .GPS. 1 u 12 64 77 28.100 0.036 0.609
+172.2*.0.53 .GPS. 1 u 15 64 77 27.023 0.089 2.204
+172.2*.4.33 .GPS. 1 u 17 64 77 27.805 0.019 8.930
172.2*.0.53 .GPS. 1 u 8 64 77 26.366 0.058 0.272
-172.2*.0.33 .GPS. 1 u 9 64 77 44.638 -3.868 0.178
+172.2*.0.33 .GPS. 1 u 6 64 77 63.905 -0.048 0.202
+172.3*.1.33 .GPS. 1 u 2 64 77 53.065 0.149 0.249
-172.3*.1.53 .GPS. 1 u 8 64 77 60.009 0.383 0.210
root@ntp02# ntpq -c rl
associd=0 status=0615 leap_none, sync_ntp, 1 event, clock_sync,
version="ntpd 4.2.6p5@1.2349-o Wed Nov 11 17:58:54 UTC 2015 (1)",
processor="x86_64", system="Linux/3.13.0-77-generic", leap=00, stratum=2,
precision=-22, rootdelay=15.249, rootdisp=12.094, refid=172.2*.8.33,
reftime=da5de23d.7bd25b22 Thu, Feb 4 2016 9:44:13.483,
clock=da5de27c.d7614550 Thu, Feb 4 2016 9:45:16.841, peer=64360, tc=6,
mintc=3, offset=-0.249, frequency=25.003, sys_jitter=1.187,
clk_jitter=0.679, clk_wander=0.032
[root@ntp03# ntpq -pn
remote refid st t when poll reach delay offset jitter
==============================================================================
*172.2*.8.33 .GPS. 1 u 20 64 377 15.426 0.198 0.102
+172.2*.0.53 .GPS. 1 u 65 64 377 27.013 0.071 0.210
#172.2*.0.33 .GPS. 1 u 2 64 377 45.691 -3.602 5.907
+172.2*.0.33 .GPS. 1 u 2 64 377 64.399 0.154 0.233
+172.3*.1.33 .GPS. 1 u 66 64 377 53.187 -0.039 0.673
-172.3*.1.53 .GPS. 1 u 65 64 377 60.901 0.290 0.199
+172.2*.64.53 .GPS. 1 u 66 64 377 19.486 -0.249 0.413
+172.2*.0.33 .GPS. 1 u 65 64 377 28.200 -0.181 0.410
+172.2*.0.53 .GPS. 1 u 65 64 377 27.064 -0.190 0.445
[root@ntp03# ntpq -c rl
associd=0 status=0615 leap_none, sync_ntp, 1 event, clock_sync,
version="ntpd 4.2.6p5@1.2349-o Mon Jan 25 14:27:34 UTC 2016 (1)",
processor="x86_64", system="Linux/3.10.0-327.4.5.el7.x86_64", leap=00,
stratum=2, precision=-24, rootdelay=15.398, rootdisp=3.084,
refid=172.2*.8.33,
reftime=da5de2f1.bbef8015 Thu, Feb 4 2016 9:47:13.734,
clock=da5de312.4aa7b5bd Thu, Feb 4 2016 9:47:46.291, peer=49919, tc=6,
mintc=3, offset=0.088, frequency=28.576, sys_jitter=0.119,
clk_jitter=0.179, clk_wander=0.033
--------------------------------------------------------------------
Thanks.
Dan
_______________________________________________
questions mailing list
questions@lists.ntp.org
http://lists.ntp.org/listinfo/questions
