Haai,

On Fri, June 29, 2007 20:15, Wraukon the Excellent wrote:
> If it gets moved into user space, you're STILL crossing from k to u,
> incurring at least double the context switches per request.

When in luser space, componentification is fairly trivial; a lot of
complexity will vanish, enabling us to both tune the newly-created d's and
the remaining parts of the kernel much more intensively with far less
effort -- obviously, the latter will somewhat compensate for the switching
deficiency.

>
> Of course with machines going faster and faster, the differences are
> becoming less perceptible and nobody cares anyway.

Exactly. I'm not a fan of lowering (or even dropping) support for
anything, but there is a time when you just have to move on. We don't
support the PDP-11, for example, even though to my knowledge there are
some running PDPs still around.

Baai,

De Zeurkous
-----------

Friggin' Machines!

>
> On 6/29/07, De Zeurkous <zeurkous@nichten.info> wrote:
>> Note: last message erroneously sent. My apologies.
>>
>> Haai,
>>
>> On Fri, June 29, 2007 14:25, Dieter Baron wrote:
>> > hi,
>> >
>> >[snip]
>> >> Your method #3 sounds pretty attractive, as it will allow new
>> >> compression
>> >> methods to be added easily. I'm not sure it would be significantly
>> >> slower.
>> >
>> >   I was afraid that crossing the kernel <-> user space barrier on each
>> > request would slow things down considerably.
>>
>> All the more reason to move more into luser space.
>>
>> Baai,
>>
>> De Zeurkous
>> -----------
>>
>> Friggin' Machines!
>>
>> >  But yes, it would be
>> > neat to have that ability, and it is by far the most flexible.
>> >
>> >
>> >                                       yours,
>> >                                       dillo
>[snip whitespace]