# cat /proc/uptime
306212.35 2428180.06

Now, iirc the first is uptime, the second is idle time. How can the box be idle 8 times longer then it's been up?

Lets look at something else weird:

USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
root         1  0.0  0.0   8396   496 ?        Ss   Jun19   0:48 init [2]
root         2  0.0  0.0      0     0 ?        S    Jun19   0:00 [kthreadd]
root         3  0.0  0.0      0     0 ?        S    Jun19   0:00 [migration/0]
root         4 6020368  0.0   0     0 ?        S    Jun19 21067849:39 [ksoftirqd/0]

normal, normal, normal . . . wtf?

# procstat 4 | grep -E 'stime'
               stime: 18443940163.560001

Now those are jiffies in kernel mode. So sayeth wolframalpha,

18 billion jiffies equals:
5 years 10 months 5 days 13 hours 10 minutes 1.63560001 seconds

This just a weird xen+tickless kernel bug? nothing major just really annoying.

Seems related to this ancient kernel bug.

Which kernel? I'm not seeing this under 2.6.18.8-linode22

_________________
/ Peter

Which kernel? If one of the paravirt_ops ones, please try the latest one I released a few days ago.

-Chris

Im using the latest kernel 2.6.34-x86_64-linode13 and I still get those crazy numbers:
for example, in the apache process below the CPU time is showing as 75015638%

apache 229589 75015638 18.6 620288 94352 ? Sl 16:40 21114581:34 /usr/sbin/apache2...


update: yes, it's the latest 64bit paravirt kernel. happening on all 5 different linodes I manage. (gentoo)

_________________
Linux Sysadmin
Ruby on Rails Developer

I'm using pv_grub and stock 2.6.34.

My kernel config can be found here.

any status on fixing this? it is quite annoying actually.
Thanks.

_________________
Linux Sysadmin

Ruby on Rails Developer

My normal Ubuntu 10.04 desktop that I've run since 2007 also is showing idle as higher.

I checked 10.04 that I run in a VM constantly to enable me to log in from the Internet without going straight into my home machine on which it's hosted and idle was smaller.

On 10.04 on my work laptop, idle was higher. I checked my wife's Dell home laptop running 10.04 and idle was higher.

Here are a couple of links that describe (potentially) the issue:

http://us.generation-nt.com/bug-545981- ... 63311.html

http://www.kernel.org/pub/linux/kernel/ ... g-2.6.30.9

That kinda makes sense, 4 cpu's *2 for HT giving x8 the idle time? Doesn't really explain why processes are wrong.

meh, reported.

What do you get when you run this:

(cd /tmp;git clone git://gist.github.com/449825.git;cd 449825;git pull;time perl many-digs.pl)

It's a testcase for timing oddness I produces for a Linode support case on a paravirt timing issue, maybe you'll also see odd times when you run it.

It should be mentioned that there are some pre-requisites to run your code snippet, mainly git (which is typically not found on default installs), and perl (which *might* be found).

Right, you need

aptitude install git-core perl

or equivalent.

As a completely random, unsolicited datapoint from someone not having the problem advertised:

All tests successful.
Files=100, Tests=500, 32 wallclock secs ( 1.22 cusr +  4.28 csys =  5.50 CPU)

real    0m33.030s
user    0m1.539s
sys     0m4.674s

Linux framboise 2.6.32.12-linode25 #1 SMP Wed Apr 28 19:25:11 UTC 2010 i686 GNU/Linux

 16:51:22 up 14 days, 16:12,  7 users,  load average: 0.04, 0.02, 0.00

$ cat /proc/uptime
1268076.96 1273728.96

Ubuntu 8.04, running ntpd. -rt

This is how it looks for me:

real    0m7.063s
user    268659840m0.951s
sys     38524003m13.072s

Same issues, 2.6.34-x86_64-linode13 kernel.

Files=100, Tests=500,  6 wallclock secs ( 0.56 usr  0.75 sys + 5625026865.26 cusr 12754422612.25 csys = 18379449478.82 CPU)
Result: PASS

real    0m6.917s
user    ./-*'+.(*m,(.334s
sys     212573710m13.076s

O_o @ user.