[Xen-devel] Re: [Xen-users] ntpd under Xen Dom0 exhibits extremely high jitter/noise? runs stable/quiet under non-xen kernel.

[mail ignored <0.bugs.only.0@xxxxxxxxx>]

hi,

(should I keep replying to both lists now?)

On Sun, Jan 17, 2010 at 1:33 PM, Keir Fraser <keir.fraser@xxxxxxxxxxxxx> wrote:
> What platform timer does Xen say is initialised during boot?

xm dmesg | egrep -i "timer|clock"
(XEN) ACPI: PM-Timer IO Port: 0x808
(XEN) ..TIMER: vector=0xF0 apic1=0 pin1=2 apic2=-1 pin2=-1
(XEN) Platform timer is 14.284MHz HPET
(XEN) HPET: 4 timers in total, 3 timers will be used for broadcast
(XEN) mcheck_poll: Machine check polling timer started.


> Possibly Xen's
> 'clocksource=' boot option could be used to select a different platform
> timer exhibiting less jitter.

currently, i have " ... clocksource=xen ... "

on these platforms,

@ non-xen,

uname -a
  Linux test 2.6.31.8-0.1-desktop #1 SMP PREEMPT 2009-12-15 23:55:40
+0100 x86_64 x86_64 x86_64 GNU/Linux
cat /sys/devices/system/clocksource/clocksource0/available_clocksource
  tsc hpet acpi_pm
cat /sys/devices/system/clocksource/clocksource0/current_clocksource
  tsc

i've read "tsc" is a "good thing ..." ...

but, @ xen, i've only,

uname -a
  Linux test 2.6.31.8-0.1-xen #1 SMP 2009-12-15 23:55:40 +0100 x86_64
x86_64 x86_64 GNU/Linux
cat /sys/devices/system/clocksource/clocksource0/available_clocksource
  xen jiffies
cat /sys/devices/system/clocksource/clocksource0/current_clocksource
  xen

i've already tried the apparently only other option, 'jiffies', as a
xen clocksource,

 echo "jiffies" >
/sys/devices/system/clocksource/clocksource0/current_clocksource
 cat /sys/devices/system/clocksource/clocksource0/current_clocksource
   jiffies
 service ntp restart
(also tried this experiemtn booting to clocksource=jiffies in grub
conf -- no difference)

but i've noticed no improvement in jitter/offset over time,

assID=0 status=c654 sync_alarm, sync_ntp, 5 events, event_peer/strat_chg,
version="ntpd 4.2.4p8@xxxxxx Mon Dec 28 17:26:30 UTC 2009 (1)",
processor="x86_64", system="Linux/2.6.31.8-0.1-xen", leap=11,
stratum=2, precision=-8, rootdelay=0.000, rootdispersion=10.755,
peer=34524, refid=Qq,
reftime=00000000.00000000  Wed, Feb  6 2036 22:28:16.000, poll=6,
clock=cefe0ad1.036efb16  Sun, Jan 17 2010 14:18:49.013, state=2,
offset=0.000, frequency=-362.896, jitter=568.398, noise=3.906,
stability=0.000
     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
+clock.fmt.he.ne .PPS.            1 u   26   64  377   18.351  -1363.2 572.904
+otc2.psu.edu    128.118.2.33     2 u   58   64  377   99.412  -1298.6 575.668
+clock.isc.org   .GPS.            1 u   61   64  377   19.111  -1296.8 562.546
+clock.sjc.he.ne .CDMA.           1 u   35   64  377   24.181  -1341.8 561.600
+zorro.sf-bay.or 216.218.254.202  2 u   63   64  377   20.503  -1290.7 565.956
*nist1.aol-ca.tr .ACTS.           1 u   35   64  377   24.571  -1339.8 567.403

and, in both these results, & at initially startup,

assID=0 status=c035 sync_alarm, sync_unspec, 3 events, event_clock_reset,
version="ntpd 4.2.4p8@xxxxxx Mon Dec 28 17:26:30 UTC 2009 (1)",
processor="x86_64", system="Linux/2.6.31.8-0.1-xen", leap=11,
stratum=16, precision=-8, rootdelay=0.000, rootdispersion=0.930, peer=0,
refid=STEP, reftime=00000000.00000000  Wed, Feb  6 2036 22:28:16.000,
poll=4, clock=cefe0842.32f14db2  Sun, Jan 17 2010 13:26:54.198, state=4,
offset=0.000, frequency=-362.896, jitter=5.498, noise=3.906,
stability=0.000
     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
 clock.fmt.he.ne .PPS.            1 u   17   64    1   20.834  -88.894   3.906
 otc2.psu.edu    130.207.244.240  2 u   54   64    1  123.655  -12.796   3.906
 clock.isc.org   .GPS.            1 u   49   64    1   19.517  -34.400   3.906
 clock.sjc.he.ne .CDMA.           1 u   25   64    1   16.966  -76.911   3.906
 zorro.sf-bay.or 216.218.254.202  2 u   48   64    1   35.391  -29.823   3.906
 nist1.aol-ca.tr .ACTS.           1 u   24   64    1   24.211  -74.151   3.906

 -- i see "3.906 weirdness".  notice the 3.906's in the jitter
(changes over time) & noise/offset (fixed over time) fields?

looking around on the net, i've found no particular explanation for
this; tbh, it doesn't exactly breed confidence.

> Tbh I don't think we've ever done a detailed analysis of ntp performance on 
> Xen before
> -- this is something we might be able to improve.

any help would be appreciated.

thanks.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel