Re: [PATCH] Shrinks sizeof(files_struct) and better layout

[Eric Dumazet <dada1@cosmosbay.com>]

David Lang a écrit :
> On Wed, 4 Jan 2006, Eric Dumazet wrote:
>> Andi Kleen a écrit :
>>> Eric Dumazet <dada1@cosmosbay.com> writes:
>>>> 1) Reduces the size of (struct fdtable) to exactly 64 bytes on 32bits
>>>> platforms, lowering kmalloc() allocated space by 50%.
>>>
>>> It should be probably a kmem_cache_alloc() instead of a kmalloc
>>> in the first place anyways. This would reduce fragmentation.
>>
>> Well in theory yes, if you really expect thousand of tasks running...
>> But for most machines, number of concurrent tasks is < 200, and using 
>> a special cache for this is not a win.
> 
> is it enough of a win on machines with thousands of concurrent tasks 
> that it would be a useful config option?

Well..., not if NR_CPUS is big too. (We just saw a thread on lkml about 
raising NR_CPUS to 1024 on ia64).

On a 1024 CPUS machine, a kmem cache could use at least 1 MB for its internal 
structures, plus 1024 pages (PAGE_SIZE) for holding the caches (one cache per 
CPU), if you assume at least one task was created on behalf each cpu.

if PAGE_SIZE is 64KB, you end up with 65 MB of ram for the cache. Even with 
100.000 tasks running on the machine, its not a win.

slab caches are very complex machinery that consume O(NR_CPUS) ram.

Eric