[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: x86 kernel flavours was: 2.6.14-1 in incoming, status and future



On Tue, Nov 01, 2005 at 11:25:37AM +0900, Horms wrote:
> 
> There was some discussion of cutting down on the number of flavours.
> It all centred around, is there any real benifit. For instance,
> it is conventional wisdom that 686 will run faster on a UP box than 686-smp,
> but for a typical workload, is it really enough to warrant an extra
> flavour.
> 
> To be honest, most of the reason for extra flavours, especially in i386,
> comes down to performance. A 686-smp kernel will run faster on a P4 than
> a 386 kernel (n.b 386 and 686 are just the flavour names). In that
> case its almost certainly worth it. But its not entirely clear there
> is enough benifit to warrant all the flavours in between.
> 
> However, as its a performance issue, what is needed is numbers.
> I heard that Ubuntu were looking into it, but haven't heard
> anything of late.

I think I may have mentioned this to you privately but there are some
people working on it, here is one approach

<http://lwn.net/Articles/160295/>

Which looks like it obviates the need, eventually, for distinct -smp and
-up builds of the kernel.  Couple that with run-time detection of cpu
type and 'apt-get install linux-image-2.6.<x>' might be all that is
eventually required.

Cheers,
Anand

Attachment: signature.asc
Description: Digital signature


Reply to: