Re: 4mb PAGES for mbuf clusters

[Michel Talon <talon@xxxxxxxxxxxxxxxx>]

km b wrote:

> On 7/13/07, Matthew Dillon <dillon@apollo.backplane.com> wrote:
>>     It's not a bad idea and tests done on FreeBSD have shown a
>>     significant improvement in performance for certain workloads,
>>     but it's fairly difficult to manage memory in 4MB chunks
>>     and fragmentation will screw it up on machines with long uptimes,
>>     so unless we rewrite a whole lot of subsystems I don't think we
>>     can do better then a hack.  For the record, I don't like what FreeBSD
>>     did.
> 
> Would you be able to recall what FreeBSD did? I couldn't find anything
> on mail archives. If it's worth I'll give it a try for private use
> even though it's a hack.
> 
> Cheers
> kmb
> 
> 

Large pages have not been integrated to FreeBSD up to now. Probably Matt 
refers to some non public perforce repositories of Alan Cox. I suppose that,
since Alan Cox announced large pages support at Christmas and they are
still not here, he is not happy with the present state of affairs. By the
way, large pages support exists in Linux since the beginning of kernel 2.6,
and perhaps before.


-- 
Michel Talon