on Rodent directory reading speed

[Auke Kok]

On 06/25/2011 04:12 AM, Daniele Guerrieri wrote:
> On Fri, Jun 24, 2011 at 9:46 PM, edscott wilson
> <edscott.wilson.garcia at gmail.com>  wrote:
>> I'm editing the subject as the mailing list guidelines indicate. (sorry I
>> did not do that sooner)
>> That is an operating system glitch. The first time you read the directory,
>> you really read the directory, while the second time, the operating system
>> has it all in memory cache and can determine that it is useless to do the
>> physical read again and will give you the cache data: much, much faster.
>> Rodent handles an additional stat cache which gemake a reload even faster. So
>> it all depends who got to go first.
>>
> I've made a better test as suggested by Auke , rebooting after each
> trial (i could also  echo 3>  /proc/sys/vm/drop_caches ? ):
>
> open /usr/bin, 2120 items, 266.7 mb:
>
> Thunar: first read: ~18s
> 	second read: ~3.6s
>
> Rodent: first read: ~24s
> 	second read: ~5.9s
>
>
> So the differences are not huge (~2s on second read)

well, the numbers speak for themselves really:

1st read: Rodent is 33% slower
2nd read: Rodent is 64% slower

Given a 2k dentry size, this seems a significant difference even in pure 
statistical terms.

Auke