scheduler

Hi,

I am trying to port a linux(2.6)(posix compliant) software based on
ARM9 to VxWorks(6.5)(Posix
Compliant) running on ARM9. The software deals with lot of
thread(posix) creation and intertask communications. It also deals
with shared memory.
I am worried that the scheduler of linux and vxworks might be
different and hence will have its impact on this porting activitiy.
I am eager to know the differences between the scheduler of linux
kernel 2.6 and Vxworks 6.5 so that i can take that into
consideration ? Has anyone analysed this or come across such a
situation ? Any links/ideas ?

Thx in advans,
Karthik Balaguru

* karthikbalaguru peremptorily fired off this memo:

> I am trying to port a linux(2.6)(posix compliant) software based on
> ARM9 to VxWorks(6.5)(Posix
> Compliant) running on ARM9. The software deals with lot of
> thread(posix) creation and intertask communications. It also deals
> with shared memory.
> I am worried that the scheduler of linux and vxworks might be
> different and hence will have its impact on this porting activitiy.
> I am eager to know the differences between the scheduler of linux
> kernel 2.6 and Vxworks 6.5 so that i can take that into
> consideration ? Has anyone analysed this or come across such a
> situation ? Any links/ideas ?

Have you contacted your support rep at vxworks?

-- 
It's a business I don't know anything about, but I admire Bill Gates
enormously. I know him individually, and I think he's incredible in business.
   -- Warren Buffett, in lecture at Kenan-Flagler Business School, University
of North Carolina at Chapel Hill. (1994)

On May 14, 4:56 pm, karthikbalaguru <karthikbalagur...@gmail.com>
wrote:
> Hi,
>
> I am trying to port a linux(2.6)(posix compliant) software based on
> ARM9 to VxWorks(6.5)(Posix
> Compliant) running on ARM9. The software deals with lot of
> thread(posix) creation and intertask communications. It also deals
> with shared memory.
> I am worried that the scheduler of linux and vxworks might be
> different and hence will have its impact on this porting activitiy.
> I am eager to know the differences between the scheduler of linux
> kernel 2.6 and Vxworks 6.5 so that i can take that into
> consideration ? Has anyone analysed this or come across such a
> situation ? Any links/ideas ?

Maybe you should study the application and figure out exactly what
operating system features are being used, and see if you can form an
idea of how sensitive the program is to the detailed behavior of those
features.

At the very least you should make some charts of the different
threads / processes and how they interact.

A lot of time it won't be very sensitive at all - but if the processor
is running flat out, differences in the scheduler or even the base
system load could break it.

On May 14, 1:56 pm, karthikbalaguru <karthikbalagur...@gmail.com>
wrote:
> Hi,
>
> I am trying to port a linux(2.6)(posix compliant) software based on
> ARM9 to VxWorks(6.5)(Posix
> Compliant) running on ARM9. The software deals with lot of
> thread(posix) creation and intertask communications. It also deals
> with shared memory.
> I am worried that the scheduler of linux and vxworks might be
> different and hence will have its impact on this porting activitiy.
> I am eager to know the differences between the scheduler of linux
> kernel 2.6 and Vxworks 6.5 so that i can take that into
> consideration ? Has anyone analysed this or come across such a
> situation ? Any links/ideas ?
>
> Thx in advans,
> Karthik Balaguru

VxWorks 6.5 (or possibly 6.6) supports full POSIX compliant scheduling
algorithms IFF you are running your code as an RTP (i.e. a process)

I'd recommend you review the documentation in the Applications
Programmers' Guide on this area. Check out the POSIX chapter.

The only additional characteristic you need to be aware of is that
threads in VxWorks are globally scheduled; by that I mean that
if you have multiple RTPs executing, each RTP doesn't get a
slice of time... it will still be the thread with the overall highest
priority in the complete system that will be scheduled to run.

Cheers!

Jason

On May 15, 8:25=A0pm, Jason <ja...@obroin.net> wrote:
> On May 14, 1:56 pm, karthikbalaguru <karthikbalagur...@gmail.com>
> wrote:
>
>
>
>
>
> > Hi,
>
> > I am trying to port a linux(2.6)(posix compliant) software based on
> > ARM9 to VxWorks(6.5)(Posix
> > Compliant) running on ARM9. The software deals with lot of
> > thread(posix) creation and intertask communications. It also deals
> > with shared memory.
> > I am worried that the scheduler of linux and vxworks might be
> > different and hence will have its impact on this porting activitiy.
> > I am eager to know the differences between the scheduler of linux
> > kernel 2.6 and Vxworks 6.5 so that i can take that into
> > consideration ? Has anyone analysed this or come across such a
> > situation ? Any links/ideas ?
>
> > Thx in advans,
> > Karthik Balaguru
>
> VxWorks 6.5 (or possibly 6.6) supports full POSIX compliant scheduling
> algorithms IFF you are running your code as an RTP (i.e. a process)
>
> I'd recommend you review the documentation in the Applications
> Programmers' Guide on this area. Check out the POSIX chapter.
>
> The only additional characteristic you need to be aware of is that
> threads in VxWorks are globally scheduled; by that I mean that
> if you have multiple RTPs executing, each RTP doesn't get a
> slice of time... it will still be the thread with the overall highest
> priority in the complete system that will be scheduled to run.
>

If process P1 has 5 runnable threads and process P2 has
3 runnable threads, assuming all the 8 threads are at same
priority, then each thread would receive one-eighth of the
CPU time.
And this will get adjusted whenever a new thread gets
created.
And so, if process P1 has 7 threads and process P2 has
3 runnable threads, assuming all the 10 threads are at the
same priority, then each thread would receive one-tenth
of the CPU time.

Can you pls tell me if VxWorks behave as stated above
with respect to threading ?

And, does VxWorks support NPTL ?
How far is VxWorks compatible with NPTL ?

Is the VxWorks scheduler similar to that of CFS
scheduler in linux ? Or
Does it use O(1) Scheduler ?

Any ideas / tips.

Thx in advans,
Karthik Balaguru

karthikbalaguru wrote:
> 
... snip ...
> 
> If process P1 has 5 runnable threads and process P2 has 3
> runnable threads, assuming all the 8 threads are at same
> priority, then each thread would receive one-eighth of the CPU
> time.  And this will get adjusted whenever a new thread gets
> created.  And so, if process P1 has 7 threads and process P2 has
> 3 runnable threads, assuming all the 10 threads are at the same
> priority, then each thread would receive one-tenth of the CPU
> time.

However, if any thread is blocked (as for i/o) it just doesn't
count until that block is lifted.

-- 
 [mail]: Chuck F (cbfalconer at maineline dot net) 
 [page]: <http://cbfalconer.home.att.net>
            Try the download section.


** Posted from http://www.teranews.com **

On May 24, 11:52=A0pm, karthikbalaguru <karthikbalagur...@gmail.com>
wrote:
> On May 15, 8:25=A0pm, Jason <ja...@obroin.net> wrote:
>
>
>
>
>
> > On May 14, 1:56 pm, karthikbalaguru <karthikbalagur...@gmail.com>
> > wrote:
>
> > > Hi,
>
> > > I am trying to port a linux(2.6)(posix compliant) software based on
> > > ARM9 to VxWorks(6.5)(Posix
> > > Compliant) running on ARM9. The software deals with lot of
> > > thread(posix) creation and intertask communications. It also deals
> > > with shared memory.
> > > I am worried that the scheduler of linux and vxworks might be
> > > different and hence will have its impact on this porting activitiy.
> > > I am eager to know the differences between the scheduler of linux
> > > kernel 2.6 and Vxworks 6.5 so that i can take that into
> > > consideration ? Has anyone analysed this or come across such a
> > > situation ? Any links/ideas ?
>
> > > Thx in advans,
> > > Karthik Balaguru
>
> > VxWorks 6.5 (or possibly 6.6) supports full POSIX compliant scheduling
> > algorithms IFF you are running your code as an RTP (i.e. a process)
>
> > I'd recommend you review the documentation in the Applications
> > Programmers' Guide on this area. Check out the POSIX chapter.
>
> > The only additional characteristic you need to be aware of is that
> > threads in VxWorks are globally scheduled; by that I mean that
> > if you have multiple RTPs executing, each RTP doesn't get a
> > slice of time... it will still be the thread with the overall highest
> > priority in the complete system that will be scheduled to run.
>
> If process P1 has 5 runnable threads and process P2 has
> 3 runnable threads, assuming all the 8 threads are at same
> priority, then each thread would receive one-eighth of the
> CPU time.
> And this will get adjusted whenever a new thread gets
> created.
> And so, if process P1 has 7 threads and process P2 has
> 3 runnable threads, assuming all the 10 threads are at the
> same priority, then each thread would receive one-tenth
> of the CPU time.
>
> Can you pls tell me if VxWorks behave as stated above
> with respect to threading ?
>

I got the below info for the above from internet ->
In VxWorks,
Round-robin scheduling Like preemptive priority-based
scheduling but it also attempts the share the CPU fairly
among all tasks of the same priority using the so called
time-slicing technique. In time-slicing each task can run
freely until its preempted by a higher priority task or its
time-slice has ended. In the latter case another equal
priority task is scheduled to run. Thus the equal-priority
task rotate, each executing for an equal interval of time.
In vxworks one can activate round-robin scheduling with
the function kernelTimeSlice() with the specified timeslice.

I find that the below link has good info w.r.t porting
between vxworks and linux.
http://www.devicegate.com/download/VxWorkstolinux.pdf

Further from the man page of pthreads i find the below
info ->

LinuxThreads-  This is the original (now obsolete)
Pthreads implementation.

NPTL (Native POSIX Threads Library)- This is the
modern Pthreads implementation. By comparison
with LinuxThreads, NPTL provides closer conformance
to the requirements of the POSIX.1 specification and
better performance when creating large numbers of
threads. NPTL requires features that are present in
the Linux 2.6 kernel.

Both of these are so-called 1:1 implementations,
meaning that each thread maps to a kernel scheduling
entity.

Modern GNU C libraries provide both LinuxThreads
and NPTL, with the latter being the default (if
supported by the underlying kernel).

But, i am unable fo find info for the below queries ->
And, does VxWorks support NPTL ?
How far is VxWorks compatible with NPTL ?

How to make the VxWorks scheduler
to act similar to that of CFS scheduler and
O(1) scheduler of linux ?

Any ideas / links ?

Thx in advans,
Karthik Balaguru

On May 24, 9:48 pm, karthikbalaguru <karthikbalagur...@gmail.com>
wrote:
> On May 24, 11:52 pm, karthikbalaguru <karthikbalagur...@gmail.com>
> wrote:
>
>
>
> > On May 15, 8:25 pm, Jason <ja...@obroin.net> wrote:
>
> > > On May 14, 1:56 pm, karthikbalaguru <karthikbalagur...@gmail.com>
> > > wrote:
>
> > > > Hi,
>
> > > > I am trying to port a linux(2.6)(posix compliant) software based on
> > > > ARM9 to VxWorks(6.5)(Posix
> > > > Compliant) running on ARM9. The software deals with lot of
> > > > thread(posix) creation and intertask communications. It also deals
> > > > with shared memory.
> > > > I am worried that the scheduler of linux and vxworks might be
> > > > different and hence will have its impact on this porting activitiy.
> > > > I am eager to know the differences between the scheduler of linux
> > > > kernel 2.6 and Vxworks 6.5 so that i can take that into
> > > > consideration ? Has anyone analysed this or come across such a
> > > > situation ? Any links/ideas ?
>
> > > > Thx in advans,
> > > > Karthik Balaguru
>
> > > VxWorks 6.5 (or possibly 6.6) supports full POSIX compliant scheduling
> > > algorithms IFF you are running your code as an RTP (i.e. a process)
>
> > > I'd recommend you review the documentation in the Applications
> > > Programmers' Guide on this area. Check out the POSIX chapter.
>
> > > The only additional characteristic you need to be aware of is that
> > > threads in VxWorks are globally scheduled; by that I mean that
> > > if you have multiple RTPs executing, each RTP doesn't get a
> > > slice of time... it will still be the thread with the overall highest
> > > priority in the complete system that will be scheduled to run.
>
> > If process P1 has 5 runnable threads and process P2 has
> > 3 runnable threads, assuming all the 8 threads are at same
> > priority, then each thread would receive one-eighth of the
> > CPU time.
> > And this will get adjusted whenever a new thread gets
> > created.
> > And so, if process P1 has 7 threads and process P2 has
> > 3 runnable threads, assuming all the 10 threads are at the
> > same priority, then each thread would receive one-tenth
> > of the CPU time.
>
> > Can you pls tell me if VxWorks behave as stated above
> > with respect to threading ?
>
> I got the below info for the above from internet ->
> In VxWorks,
> Round-robin scheduling Like preemptive priority-based
> scheduling but it also attempts the share the CPU fairly
> among all tasks of the same priority using the so called
> time-slicing technique. In time-slicing each task can run
> freely until its preempted by a higher priority task or its
> time-slice has ended. In the latter case another equal
> priority task is scheduled to run. Thus the equal-priority
> task rotate, each executing for an equal interval of time.
> In vxworks one can activate round-robin scheduling with
> the function kernelTimeSlice() with the specified timeslice.
>
> I find that the below link has good info w.r.t porting
> between vxworks and linux.http://www.devicegate.com/download/VxWorkstolinux.pdf
>
> Further from the man page of pthreads i find the below
> info ->
>
> LinuxThreads-  This is the original (now obsolete)
> Pthreads implementation.
>
> NPTL (Native POSIX Threads Library)- This is the
> modern Pthreads implementation. By comparison
> with LinuxThreads, NPTL provides closer conformance
> to the requirements of the POSIX.1 specification and
> better performance when creating large numbers of
> threads. NPTL requires features that are present in
> the Linux 2.6 kernel.
>
> Both of these are so-called 1:1 implementations,
> meaning that each thread maps to a kernel scheduling
> entity.
>
> Modern GNU C libraries provide both LinuxThreads
> and NPTL, with the latter being the default (if
> supported by the underlying kernel).
>
> But, i am unable fo find info for the below queries ->
> And, does VxWorks support NPTL ?
> How far is VxWorks compatible with NPTL ?
>
> How to make the VxWorks scheduler
> to act similar to that of CFS scheduler and
> O(1) scheduler of linux ?
>
> Any ideas / links ?
>
> Thx in advans,
> Karthik Balaguru

The default scheduler is a priority based scheduler; in the example
you gave, the various threads in the two processes at the same
thread priority will not necessarily get equal processing time.

For example, if one of the threads did

for (;;) {}

it would continue to hold the CPU indefinitely unless a higher
priority event caused a reschedule, and even then it would
regain the CPU because it is still at the top of the ready to
run list.

If the thread blocked (because it yielded, or waited on a resource
such as a semaphore), another thread at the same priority
could run.

As you noted, you can turn on time base "round-robin"
scheduling which will limit the amount of time any thing
within a particularly priority level gets to run.

VxWorks 6.2 (might be 6.3, I'm not sure), added full
POSIX thread scheduling capabilities within an
RTP ("real time process"), although I'm not familiar
enough with this algorithm to know how it works if
you have multiple processes running.

There's detailed documentation on the scheduling
algorithms in the Application' Programmer's Guide

Cheers!

Jason