Re: kernel-{image,headers} package bloat
On 04/30/2001 03:21:55 PM "Christopher C. Chimelis" wrote:
>> Basically, I can understand everyone's desires for a kernel that covers
>> their cases (SMP, UP, 686, 386, etc), but the bloat issue that initially
I can't understand that desire for such a small gain, but whatever.
>> this thread gets more and more Intel/AMD-centric, I'm beginning to
wonder
>> what the larger implications may be...
I would think an even scarier larger implication is that on my desktop
machines, it's not unusual to have less than 10% kernel utilization. Thus
even in the wildest dreams, no level of "optimization" can have more than a
2% effect on overall CPU speed. But what happens when people provide /
request / demand optimized versions of Gnome, KDE, and Apache for every
possible combination of optimization? I'm sure an optimized libc would
have more impact that an optimized kernel. I'm sure my webserver would
benefit far more from "optimized" Apache and Perl than it would from an
"optimized" kernel, based on crude observations of the "top" command.
The problem is not providing "optimized" versions of one "little" package,
the problem is opening the floodgates to recompiles of every package in
Debian with every possible combination of optimization.
In an example that is fairly silly, if I recompile the kernel for a machine
with a CPU that costs $300 (my processors are usually cheaper) and it gets
a 20% kernel performance boost (totally unrealistic daydream) and it
typically spends 10% of it's time in kernel space (probably less), that's
the equivalent of saving me $6 over the cost of just buying the next faster
processor. Now compare $6 to the cost of the electricity to run the box,
or my salary to properly install and test the "optimized" kernel and I just
don't see the payoff. Then there are the non-monetary costs, someone
spending time making 20 different compilations of the same package is not
spending time fixing bugs or playing Quake or drinking beer.
Or on my firewall and kiosk machines using $50 special computers, the
processor is probably worth $5 so I'd figure the optimizations save me 10
cents over the cost of upgrading the CPU to a faster one.
Reply to: