On Wed, Oct 22, 2008 at 2:04 PM, Frédéric Weisbecker <fweisbec@gmail.com> wrote:
quoted text
> 2008/10/22 sbs <gexlie@gmail.com>:
>> after upgrading from 2.6.16 to 2.6.27 kernel top shows a strange
>> picture of extremely highly %CPU values.
>>
>> # top -V
>>        top: procps version 3.2.6
>>
>>
>> # top
>> top - 04:24:06 up 7 days,  1:12,  1 user,  load average: 2.28, 2.03, 2.00
>> Tasks: 183 total,   2 running, 181 sleeping,   0 stopped,   0 zombie
>> Cpu(s): 36.2% us,  2.4% sy,  0.0% ni, 56.1% id,  4.7% wa,  0.1% hi,  0.5% si
>> Mem:  12473092k total, 11169060k used,  1304032k free,       16k buffers
>> Swap:  2048248k total,      636k used,  2047612k free,  1257988k cached
>>
>>  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
>> 19824 b      20   0 92212  17m 5420 S 4308  0.1   1:39.49 httpd
>> 14396 b      20   0 92716  28m  16m S 3721  0.2  38:47.38 httpd
>> 14024 b      20   0 92376  28m  16m S 2437  0.2  51:49.68 httpd
>>  7415 b      20   0 96876  40m  24m D 2146  0.3  93:59.72 httpd
>>  8182 b      20   0 94500  36m  21m S 1997  0.3 111:41.60 httpd
>> 13461 b      20   0 94240  31m  17m S 1439  0.3  48:43.71 httpd
>>  8466 b      20   0 92036  32m  21m S 1136  0.3 107:37.90 httpd
>> 17620 b      20   0 92660  23m  11m S 1003  0.2  10:55.03 httpd
>> 14025 b      20   0 92000  28m  16m S  720  0.2  51:49.49 httpd
>>  3808 nobody    20   0 3013m 2.9g  592 S  714 24.7 343030:50 memcached
>>   10 root     15  -5     0    0    0 S  714  0.0  11881:28 ksoftirqd/3
>> 18487 b      20   0 91416  20m 9572 S  580  0.2   8:29.20 httpd
>> 13252 b      20   0 93408  30m  17m S  580  0.3  45:19.69 httpd
>> 17525 b      20   0 93208  23m  10m S  566  0.2  18:34.40 httpd
>> 14760 b      20   0 93452  28m  15m S  534  0.2  57:34.65 httpd
>>  3804 nobody    20   0 2995m 2.9g  592 S  441 24.6 378908:26 memcached
>> 12209 b      20   0 92312  28m  16m S  279  0.2  63:28.45 httpd
>>
>>
>>
>> %CPU is 4308 here and it's not the highest value.
>> i havent seen that on any previous kernel on the same server.
>>
>> should i upgrade procps or downgrade kernel then to get proper values back? :)
>> or is it a feature of the new kernel?
>>
>> thank you
>>
>
> Hi,
>
> It really depends on what your box is doing at the time you capture
> the cpu usage.
> And it really depends too of your kernel config...

It really shouldn't, we're talking about /proc which is supposed to be
a stable userspace interface.

Try upgrading your procps, there's a chance it's just so ancient that
it accidentally got broken. However, if that fixes your problem, then
please report back with a subject line of "/proc API breakage" or
somesuch, to get the proper attention to the issue.