[wellylug] Sarge approaching release

[David Antliff]

On Sat, 4 Sep 2004, Enkidu wrote:
> I disagree with the "many application". A few esoteric ones, maybe.

Anything that processes volumes of data like video, sound, games in Linux.
The difference is significant for these sorts of applications. You may not
run many of them, but lots of people do. Software decoding of MPEG is a
good example.

> >A kernel "built for your CPU" doesn't really offer much advantage for
> >'home' users - like any decent OS, Linux doesn't spend a significant
> >amount of time running in the kernel, so many optimisations offer little
> >improvement to a desktop PC. A heavily loaded host servicing many
> >connections at once will benefit more.
> >
> Um, no, not really. It would need network optimisation, not kernel
> optimisation.

Network optimisation - what exactly is that? The kernel is accessed by any
user program via system calls. Heavy network-utilising programs make lots
of system calls (read, write mostly). Optimising the time the kernel
spends processing system calls has a direct effect on throughput. Context
switches take time too. Heavy network load can lead to heavy use of system
calls. However most system calls simply block on IO so optimisation there
isn't helpful. What i'm trying to say is that optimising your kernel
doesn't provide much benefit for typical operations - I suspect we're
arguing the same point here.

BTW when I said "running in the kernel" previously, I was excluding the
time spent in the kernel waiting for hardware.

> What sort of reasons have you found that need kernel compiles? Most of
> the compiles that I've had to do have been for reasons *other* than
> speed.

I didn't say I compile my kernel for speed. I actually implied there's
little point for a desktop situation. As I said, I think we're arguing the
same point.

-- 
David.