lkml.org 
[lkml]   [2018]   [Feb]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] of: cache phandle nodes to decrease cost of of_find_node_by_phandle()
From
Date
On 02/07/18 04:44, Chintan Pandya wrote:
>
>
> On 2/5/2018 5:53 PM, Chintan Pandya wrote:
>>
>>>
>>> My question was trying to determine whether the numbers reported above
>>> are for a debug configuration or a production configuration.
>> My reported numbers are from debug configuration.
>>
>>> not a production configuration, I was requesting the numbers for a
>>> production configuration.
>> I'm working on it. But please expect some delay in my response for this. As I mentioned earlier, I need to work with few teams to get these numbers.
>>
>>
>>> show a significant boot time reduction from the patch then there is
>>> less justification for adding complexity to the existing code.  I
>>> prefer to use simpler data structures and algorithms __if__ extra
>>> complexity does not provide any advantage.  The balance between
>>> complexity and benefits is a core software engineering issue.
>>>
>> Ok
>
> Avg Kernel boot time comparison in production set up:
>
> [0] Base: 4519ms
> [1] 4115ms (~400ms improvement)
> [2] 4115ms (~400ms improvement)
> [3] 4177ms (~340ms improvement)
>
> Full data:
> [1] 1024 sized pre-populated cache
> ITR-1    ITR-2    ITR-3    ITR-4    Avg
> 4115    4123    4124    4107    4115
>
> [2] Dynamic sized cache allocation/free
> ITR-1    ITR-2    ITR-3    ITR-4    Avg
> 4122    4131    4106    4118    4115
>
> [3] Fixed 64 sized cache
> ITR-1    ITR-2    ITR-3    ITR-4    Avg
> 4153    4186    4198    4181    4177
>
>
> [1] is my experimental patch and dirty enough to not get merged anywhere. So, I will not push it.
>
>
>
> Chintan

Thank you very much. This looks like a real improvement to me.

I'll rebase my patches, updated to address the comments, on 4.16-rc1.

-Frank

\
 
 \ /
  Last update: 2018-02-07 21:10    [W:0.045 / U:4.272 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site