[ros-kernel] Process scheduler / timer resolution
KJK::Hyperion
noog at libero.it
Sat Mar 13 03:15:42 CET 2004
At 02.15 13/03/2004, you wrote:
>>>Threads have been available in Linux for a looong time,
>>sure. Only they weren't threads. If not for IBM, they would probably
>>still not be there
>You'll have to explain a bit further here I think.
by "threads" I mean "NPTL", a patch that was developed by IBM and
integrated officially in 2.6. I have a hard time swallowing the concept
that clone tasks are threads
>>>PnP is not a new feature either
>>suuure. USB hotplugging in 2.4 is a walk in the park</irony>
>I don't see how USB hotplugging has to do with PnP
PnP is all about hotplugging - detecting new hardware immediately, loading
the drivers and, especially, making the devices immediately available.
Microsoft took it to an insane extreme, applying it to buses that don't
hotplug at all, like serial and parallel ports, leading to "hey dude,
where's my modem?" syndrome, but the Windows implementation is otherwise
sensible
>PnP, if my memory is correct is some "annoying feature" in some ISA
>extension cards.
never heard about it. I'm not even sure of what does "ISA" mean
>But, if you wish to call everything that works when it's plugged PnP, I'll
>go by that. Linux has had PCI support for times immemorial ;)
I think hotplugging PCI cards involves burning your power supply
>>I'm not fair. I have opinions, and they aren't objective. Backed by facts
>>yes, objective not necessarily
>Well usually one tries to have their opinions as much acurate as possible,
>at least I hope so.
the only accuracy in opinions is in how they reflect what one thinks
>Don't know too much about databases. But as long as the job gets done, and
>efficiently I don't see where is the problem.
nobody sees the problem until it's solved. Linux has had a scheduler that
ran in linear time for almost ten years, the whole time hearing from
zealots how Linux was the bestest thing since sliced bread. Now that we
have constant time scheduling, we realize Linux *was* the best... for
running a basement Apache server. The point I'm trying to make is that the
success of a system is completely unrelated to the quality of the kernel
(which has sad double-edged consequences, for us too), and history proved
this again and again. Ironically, the Windows NT line is a perfect example
for this: a great kernel being denied any sort of significant innovation
(most additions are patches over pathological issues found in previous
versions, like the global cancel spinlock, or the inefficency of event
objects as the wait object for critical sections) and forced to run some
painfully bad user-mode code. I would never, ever administer a Windows
server. The mess they've made (and keep making worse) is just too much for
me to handle
>And NTFS is not so great either,
care to elaborate?
>What are you referring to when you talk of compatibility anyway?
I didn't talk of compatibility
More information about the Ros-kernel
mailing list