- Joined
- 24 Jan 2002
- Messages
- 12,391
lots of miss information concerning the system idle process, so, for members of ntfs, I'll give you the skinny...you'll definitely have a heads up on this much not understood feature of xp
if you'd like to do more research, the Microsoft publication "inside windows 2000" is an unbelievable wealth of information and reference
everyone knows, Windows NT is multi threading operating system, which in everyday language just means you can do a bunch of different things on your box, with the illusion that everything is happening simultanously...an amazing ability this is indeed, and everyone should learn to take advantage of it...do not follow that old advice and force your computer to run one only thing at a time, your computer allows you to be so much more productive, it's designed to run many things at once...but that's another discussion.
the reality is, your standard chip can only do one thing at a time, but, it switches back and forth between threads so fast, and so often, that everything looks like it virtually (I love that word) happens at once.
this switching back and forth is managed with a set of algorithms known as "the NT scheduler".
the scheduler simply looks at the priorities on the various threads that are active...it then gives each thread it's share of CPU time, ( this unit of time is known as a quantum )...the quantum length can be adjusted in NT, but that's the word we use when speaking of the time unit the CPU spends with any given thread. Quantum lengths are also devided, and a thread will be preempted and forced to give it up it's remaining quantum immediatly if a higher priority shows up.
the scheduler doesn't look at what process a thread belongs to, it just looks at the priority of a thread...pretty simple...the scheduler then executes the highest priority...this high priority thread will decay a level for every quantum completed...however, the level can be re bumped depending on the program and how it's referenced by the user....for instance as an extreme example, a keyboard stroke can at times bump a priority on an idle thread that is waiting for that stroke by 6 points.
The scheduler will be able to prevent threads from starving due to a priority lower then everything else, (When real time threads are not ready to execute) and the scheduler will make sure every thread gets it's share of CPU...lower priority threads are bumped sometimes to level 15... they can even be given more then one quanta if they haven't had a quantum for too long a period.
there are also other policies that insure lower priorities get a quantum (when real time threads are not ready to execute)... as mentioned, every thread in the 1-15 range will decay a level (to a point set by the program) every time it completes a quantum...and of course, these threads can always get bumped up levels according to use, and so refreshed in priority as well as decayed.
programs also voluntarily give up their priority, and programs get "blocked "from even being considered for a quantum when they are not ready to execute.
"not ready to execute", or "blocked" threads use no CPU time...they incur no CPU overhead and there is no CPU management overhead, the overhead of the NT scheduler is a constant set of algorithms that always run....when a thread or process incurs zero CPU, it's as if it's not even running at all as far as CPU resource liability
if a number of threads have the same priority, these threads are just executed in a round robin fashion along that respective priority.
of course, there's allot of sophistication involved with these scheduling policies, but you now have an accurate model
fyi... even most people that know about CPU scheduling in NT are under the impression that there are 31 priorities in the NT kernel
that's incorrect
there are 32
there is 1 through 15 for the common apps, 16 through 31 for critical threads, and then there is a priority "zero"
in NT, the higher the number a priority, the higher the priority...this is a random choice, and similar to the vms's OS's, (after all, it is David cutler that is responsible for nt)...the reverse could have been true, it's just the way Microsoft did it.
as told, priorities 1 through 15 are the dynamic priorities that typical applications get assigned..the threads in this realm change priorities all along, getting bumped and degraded according to need by the os...these are dynamic in nature.
16 through 31 not dynamic, this is what we mean by "real time" when talking about a threads priority...the priority of real time threads is constant...the os does not change the priorities of these threads..these priorities are used for operations that are time critical...in addition, users with administrative authority can also set real time priorities for applications on their own
so now I finally get to the next level of priority
this is priority level 0 (zero)
your system idle process gets a priority of zero.
This process is a single thread running on each processor, which has the sole task of accounting for processor time when the system isn't processing other threads. In Task Manager, expect this process to account for the majority of processor time.
with priority zero, when no other thread is executing, priority zero gets a quantum.
and so, here is the biggest misinterpretation with the system idle process
if you open your task manager while the CPU is hardly working, and look to processes, you will notice that your system idle process is taking 90% or such of the CPU time.
HA
THIS FIGURE IS MISUNDERSTOOD!!!!
this figure represents how much cpu time has not even been requested by anything you are currently working with
if you'd like to do more research, the Microsoft publication "inside windows 2000" is an unbelievable wealth of information and reference
everyone knows, Windows NT is multi threading operating system, which in everyday language just means you can do a bunch of different things on your box, with the illusion that everything is happening simultanously...an amazing ability this is indeed, and everyone should learn to take advantage of it...do not follow that old advice and force your computer to run one only thing at a time, your computer allows you to be so much more productive, it's designed to run many things at once...but that's another discussion.
the reality is, your standard chip can only do one thing at a time, but, it switches back and forth between threads so fast, and so often, that everything looks like it virtually (I love that word) happens at once.
this switching back and forth is managed with a set of algorithms known as "the NT scheduler".
the scheduler simply looks at the priorities on the various threads that are active...it then gives each thread it's share of CPU time, ( this unit of time is known as a quantum )...the quantum length can be adjusted in NT, but that's the word we use when speaking of the time unit the CPU spends with any given thread. Quantum lengths are also devided, and a thread will be preempted and forced to give it up it's remaining quantum immediatly if a higher priority shows up.
the scheduler doesn't look at what process a thread belongs to, it just looks at the priority of a thread...pretty simple...the scheduler then executes the highest priority...this high priority thread will decay a level for every quantum completed...however, the level can be re bumped depending on the program and how it's referenced by the user....for instance as an extreme example, a keyboard stroke can at times bump a priority on an idle thread that is waiting for that stroke by 6 points.
The scheduler will be able to prevent threads from starving due to a priority lower then everything else, (When real time threads are not ready to execute) and the scheduler will make sure every thread gets it's share of CPU...lower priority threads are bumped sometimes to level 15... they can even be given more then one quanta if they haven't had a quantum for too long a period.
there are also other policies that insure lower priorities get a quantum (when real time threads are not ready to execute)... as mentioned, every thread in the 1-15 range will decay a level (to a point set by the program) every time it completes a quantum...and of course, these threads can always get bumped up levels according to use, and so refreshed in priority as well as decayed.
programs also voluntarily give up their priority, and programs get "blocked "from even being considered for a quantum when they are not ready to execute.
"not ready to execute", or "blocked" threads use no CPU time...they incur no CPU overhead and there is no CPU management overhead, the overhead of the NT scheduler is a constant set of algorithms that always run....when a thread or process incurs zero CPU, it's as if it's not even running at all as far as CPU resource liability
if a number of threads have the same priority, these threads are just executed in a round robin fashion along that respective priority.
of course, there's allot of sophistication involved with these scheduling policies, but you now have an accurate model
fyi... even most people that know about CPU scheduling in NT are under the impression that there are 31 priorities in the NT kernel
that's incorrect
there are 32
there is 1 through 15 for the common apps, 16 through 31 for critical threads, and then there is a priority "zero"
in NT, the higher the number a priority, the higher the priority...this is a random choice, and similar to the vms's OS's, (after all, it is David cutler that is responsible for nt)...the reverse could have been true, it's just the way Microsoft did it.
as told, priorities 1 through 15 are the dynamic priorities that typical applications get assigned..the threads in this realm change priorities all along, getting bumped and degraded according to need by the os...these are dynamic in nature.
16 through 31 not dynamic, this is what we mean by "real time" when talking about a threads priority...the priority of real time threads is constant...the os does not change the priorities of these threads..these priorities are used for operations that are time critical...in addition, users with administrative authority can also set real time priorities for applications on their own
so now I finally get to the next level of priority
this is priority level 0 (zero)
your system idle process gets a priority of zero.
This process is a single thread running on each processor, which has the sole task of accounting for processor time when the system isn't processing other threads. In Task Manager, expect this process to account for the majority of processor time.
with priority zero, when no other thread is executing, priority zero gets a quantum.
and so, here is the biggest misinterpretation with the system idle process
if you open your task manager while the CPU is hardly working, and look to processes, you will notice that your system idle process is taking 90% or such of the CPU time.
HA
THIS FIGURE IS MISUNDERSTOOD!!!!
this figure represents how much cpu time has not even been requested by anything you are currently working with
