On Thu, 24 Jul 2008 09:21:31 +0200, scott wrote:

>> No, I'd say that it's because Linux is a bit more efficient in its use
>> of the processor.  I wouldn't expect to see 100% utilization unless I
>> was actually running a highly intensive application like POV-Ray that
>> actually needed the CPU.  Launching a program is kid's stuff - no
>> reason for that to suck 100% of the CPU.
> 
> Of course there is, if a certain amount of code needs to be executed by
> the CPU to "launch" the program, I want my OS to use 100% CPU until it's
> done. Whether that takes 0.01ms or 1 second, it should still use 100%
> CPU until it's done.  What on Earth is the reason not to?

If that program takes 100% of the CPU, then *nothing* else gets any CPU - 
other applications, the OS, etc.  Those things need resources, too.

Jim

Post a reply to this message

Jim Henderson <nos### [at] nospamcom> wrote:
> If that program takes 100% of the CPU, then *nothing* else gets any CPU - 
> other applications, the OS, etc.  Those things need resources, too.

  That's not how pre-emptive multitasking works.

  You don't need "free CPU time" for task switching to work. It's not like
a program using 100% of CPU time would mean the OS has lost control.

  100% CPU time simply means that the OS is giving all the CPU time to
that process because nothing else needs the CPU. There's no need to keep
the CPU idle for no reason if only one process requests CPU time.

  CPUs support this thing called a clock interrupt. This means that at
certain intervals (the interval can be set up by the OS) a clock circuit
sends a signal to the CPU, which causes the CPU to stop whatever it was
doing and jump to a predefined location (which also can be set up by
the OS). This predefined location is usually the task manager of the OS.

  The task manager has a list of processes, categorized by their priority,
whether they are idle or not (there exist OS system calls which a process
can call to make itself idle) and other details.

  If only one process is non-idle, the task manager simply resumes the
execution of that process. The effective result of this is that that process
gets 100% of CPU time (well, more precisely something like 99.9999% because
the task manager takes that 0.0001% to check the list of processes).

  It's impossible for a process to block a pre-emptive OS completely.
When the clock signal interrupts the CPU, the CPU just stops executing
that process, period. There's no way around it. When that happens, the
OS is free to do whatever it wants. If another process has just started,
the OS can share CPU time with it, no problem. If they have equal priorities
and both are CPU-intensive, they will get 50% each.

  Some people really do seem to have some kind of confusion with this.
They believe that taking 100% of CPU time means that no CPU time is left
for the OS to perform task switching. This is just not how CPUs and
pre-emptive multitasking works. The OS is not hindered in any way.

-- 
                                                          - Warp

Post a reply to this message

>> Of course there is, if a certain amount of code needs to be executed by
>> the CPU to "launch" the program, I want my OS to use 100% CPU until it's
>> done. Whether that takes 0.01ms or 1 second, it should still use 100%
>> CPU until it's done.  What on Earth is the reason not to?
>
> If that program takes 100% of the CPU, then *nothing* else gets any CPU -
> other applications, the OS, etc.  Those things need resources, too.

I meant the CPU being at 100% during the program loading, not that 
particular program using 100% all for itself.  Of course I assume the OS can 
do important things, but still the CPU should not be below 100% while I am 
waiting for a program to load (assuming it's CPU bound).  That would be just 
wasted time.

Post a reply to this message

On Thu, 24 Jul 2008 11:20:54 -0400, Warp wrote:

>   That's not how pre-emptive multitasking works.

Yes, I am aware of how pre-emptive vs. non-preemptive multitasking 
works.  Been working with OSes for many, many years.

And that kinda is my point to Scott.

Jim

Post a reply to this message

On Thu, 24 Jul 2008 17:21:35 +0200, scott wrote:

>>> Of course there is, if a certain amount of code needs to be executed
>>> by the CPU to "launch" the program, I want my OS to use 100% CPU until
>>> it's done. Whether that takes 0.01ms or 1 second, it should still use
>>> 100% CPU until it's done.  What on Earth is the reason not to?
>>
>> If that program takes 100% of the CPU, then *nothing* else gets any CPU
>> - other applications, the OS, etc.  Those things need resources, too.
> 
> I meant the CPU being at 100% during the program loading, not that
> particular program using 100% all for itself.  Of course I assume the OS
> can do important things, but still the CPU should not be below 100%
> while I am waiting for a program to load (assuming it's CPU bound). 
> That would be just wasted time.

Generally that isn't going to happen on modern processors, though, 
because the CPU hits 100% during computationally intensive periods, but 
internal clock speeds on CPUs these days tend to be *several times* the 
speed of even the bus to the CPU.

One of the reasons Intel entered the networking components market was 
because CPU utilization on their faster generations of processors was 
relatively low because of the speed differential between the time it took 
to process data and the time it took to get the data (or code, rather) to 
the processor to actually process things.

That may be what you're talking about here.  But this is why you won't 
see a CPU hit 100% during program load - the bus just isn't fast enough 
relative to the speed of the CPU.  Therefore, if you want to see 100% 
load on the CPU during a program load, you need a slower processor (until 
bus speeds catch up with CPU speeds, and that won't happen because of the 
physical distance the data has to travel on the bus to get to the CPU, 
where everything is nm apart instead of inches apart.

Jim

Post a reply to this message

scott <sco### [at] scottcom> wrote:
> I meant the CPU being at 100% during the program loading, not that 
> particular program using 100% all for itself.  Of course I assume the OS can 
> do important things, but still the CPU should not be below 100% while I am 
> waiting for a program to load (assuming it's CPU bound).  That would be just 
> wasted time.

  Often when a program loads, it reads a lot of data from the disk. This
means that most programs are I/O-intensive during loading, not CPU-intensive.
(Basically the CPU has to constantly wait for the superslow hard disk to do
its job.)

  Some programs might perform some other forms of I/O during loading,
which may also cause the CPU to have to wait mostly idle. For example
a game could be loading tons of textures to the GPU memory, which takes
its time.

-- 
                                                          - Warp

Post a reply to this message

Jim Henderson <nos### [at] nospamcom> wrote:
> On Thu, 24 Jul 2008 11:20:54 -0400, Warp wrote:

> >   That's not how pre-emptive multitasking works.

> Yes, I am aware of how pre-emptive vs. non-preemptive multitasking 
> works.

  In fact, even in a cooperative multitasking system a process taking
100% of CPU time doesn't necessarily mean the system is locked into
that single process. The process may still be nicely sharing time, but
if it's the only CPU-intensive process running, it may be getting all
of the available CPU time.

-- 
                                                          - Warp

Post a reply to this message

On Thu, 24 Jul 2008 11:45:16 -0400, Warp wrote:

> Jim Henderson <nos### [at] nospamcom> wrote:
>> On Thu, 24 Jul 2008 11:20:54 -0400, Warp wrote:
> 
>> >   That's not how pre-emptive multitasking works.
> 
>> Yes, I am aware of how pre-emptive vs. non-preemptive multitasking
>> works.
> 
>   In fact, even in a cooperative multitasking system a process taking
> 100% of CPU time doesn't necessarily mean the system is locked into that
> single process. The process may still be nicely sharing time, but if
> it's the only CPU-intensive process running, it may be getting all of
> the available CPU time.

Yep, absolutely - utilization figures are typically trended, and there 
usually is more than one process in the "task queue", even when something 
tries to take over the system entirely.  Interrupts and whatnot being in 
place means there's always something else that can take control unless 
the program has the ability (at the kernel level) to truly take 
everything over.

Jim

Post a reply to this message

> Generally that isn't going to happen on modern processors, though,
> because the CPU hits 100% during computationally intensive periods,

Yeh that's what I meant, like if the CPU had to decompress some files during 
loading (eg game graphics or whatever), assuming it is CPU-bound then of 
course I want the CPU usage to be at 100%.

Post a reply to this message

On Fri, 25 Jul 2008 09:02:08 +0200, scott wrote:

>> Generally that isn't going to happen on modern processors, though,
>> because the CPU hits 100% during computationally intensive periods,
> 
> Yeh that's what I meant, like if the CPU had to decompress some files
> during loading (eg game graphics or whatever), assuming it is CPU-bound
> then of course I want the CPU usage to be at 100%.

That's a big assumption, though, and not generally the way things work.  
And even then, you're assuming that the speed of the CPU core = the speed 
of the bus to the memory, and that just isn't happening on modern systems 
any more.

Jim

Post a reply to this message