Timing comparisons for multi-kprobe feature

[Hien Nguyen]

Please ignorethe previous messages. I took the wrong calculation

(stop- start) / cycles is always one. Doh, stupid me!

Hien.
Hien Nguyen wrote:

> BTW, my machine is a 4 way 700Mhz pentium III
>
> And the same go for  multi-probe test,  here a portion of the output 
> for multi-probe test (1 microsecond)
>
> Apr 18 14:58:46 elm3b99 kernel: kprobe overhead of 10 cycles per 
> iteration
> Apr 18 14:58:46 elm3b99 kernel: PASSED: for 5
> Apr 18 14:58:46 elm3b99 kernel: kprobe start 2666398510855, stop 
> 2666398521721
> Apr 18 14:58:46 elm3b99 kernel: kprobe overhead count 10, number of 
> cycles 10866
> Apr 18 14:58:46 elm3b99 kernel: kprobe overhead of 1086 cycles per 
> iteration
> Apr 18 14:58:46 elm3b99 kernel: kprobe start 2666398548107, stop 
> 2666398558376
> Apr 18 14:58:46 elm3b99 kernel: kprobe overhead count 10, number of 
> cycles 10269
> Apr 18 14:58:46 elm3b99 kernel: kprobe overhead of 1026 cycles per 
> iteration
> Apr 18 14:58:46 elm3b99 kernel: kprobe start 2666398583828, stop 
> 2666398594476
> Apr 18 14:58:46 elm3b99 kernel: kprobe overhead count 10, number of 
> cycles 10648
> Apr 18 14:58:46 elm3b99 kernel: kprobe overhead of 1064 cycles per 
> iteration
> Apr 18 14:58:46 elm3b99 kernel: kprobe start 2666398623607, stop 
> 2666398623801
> Apr 18 14:58:47 elm3b99 kernel: kprobe overhead count 10, number of 
> cycles 194
> Apr 18 14:58:47 elm3b99 kernel: kprobe overhead of 19 cycles per 
> iteration
> Apr 18 14:58:47 elm3b99 kernel: kprobe start 2666398659568, stop 
> 2666398659686
> Apr 18 14:58:47 elm3b99 kernel: kprobe overhead count 10, number of 
> cycles 118
> Apr 18 14:58:47 elm3b99 kernel: kprobe overhead of 11 cycles per 
> iteration
> Apr 18 14:58:47 elm3b99 kernel: kprobe start 2666398683675, stop 
> 2666398683779
> Apr 18 14:58:47 elm3b99 kernel: kprobe overhead count 10, number of 
> cycles 104
> Apr 18 14:58:47 elm3b99 kernel: kprobe overhead of 10 cycles per 
> iteration
> Apr 18 14:58:47 elm3b99 kernel: PASSED: for 6
> Apr 18 14:58:47 elm3b99 kernel: kprobe start 2666398721825, stop 
> 2666398733128
>
>
> Hien Nguyen wrote:
>
>> Hi Ananth, Will,
>>
>> I can't explain this but this the output for my timing_kprobe, which 
>> my calculation yeilds 1 for both kprobe and jprobe
>>
>> Apr 18 15:58:02 elm3b99 kernel: kprobe registered
>> Apr 18 15:58:02 elm3b99 kernel: plant jprobe at f8a002b0, handler 
>> addr f8a00030
>> Apr 18 15:58:12 elm3b99 kernel: kprobe start 5169729839300, stop 
>> 5170565252311
>> Apr 18 15:58:12 elm3b99 kernel: kprobe overhead count 1000000, number 
>> of cycles 835413011
>> Apr 18 15:58:14 elm3b99 kernel: jprobe start 5170565271260, stop 
>> 5171945384508
>> Apr 18 15:58:14 elm3b99 kernel: jprobe overhead count 1000000, number 
>> of cycles 1380113248
>> Apr 18 15:58:15 elm3b99 kernel: kprobe start 5171945399260, stop 
>> 5172780564338
>> Apr 18 15:58:15 elm3b99 kernel: kprobe overhead count 1000000, number 
>> of cycles 835165078
>> Apr 18 15:58:17 elm3b99 kernel: jprobe start 5172780578137, stop 
>> 5174156636912
>> Apr 18 15:58:17 elm3b99 kernel: jprobe overhead count 1000000, number 
>> of cycles 1376058775
>> Apr 18 15:58:17 elm3b99 kernel: kprobe unregistered
>> Apr 18 15:58:17 elm3b99 kernel: jprobe unregistered
>>
>>
>> Ananth N Mavinakayanahalli wrote:
>>
>>> Hi,
>>>
>>> Here is the result of the tests I carried out to determine the cost
>>> of the multiple-kprobes feature. All tests were carried out on a
>>> uni PIII running at 732MHz.
>>>
>>> I used Will's timing_kprobe and kprobe_multi modules on 
>>> linux-2.6.12-rc2
>>> for the tests.
>>>
>>> (All values in microseconds)
>>>
>>>         Vanilla        Single        Multiple
>>>         (2.6.12-rc2)    Interface    Interfaces
>>> ------------------------------------------------------------
>>> kprobe        1.124        1.149        1.149
>>> jprobe        1.891        1.885        1.885
>>> kprobe_multi    NA        1.306        1.387
>>>
>>> Note:
>>> - Single Interface => linux-2.6.12-rc2 + take3 patch
>>> - Multiple Interfaces => linux-2.6.12-rc2 + take4 patch
>>> - kprobe_multi => averaged values with Will's kprobe_multi testcase
>>>
>>> Thanks,
>>> Ananth
>>>
>>
>>
>
>