[ECOS] ECOS Real-Time performance

[ECOS] ECOS Real-Time performance

[Meulendijks, J.]

Hi everybody,

I'm trying to choose a RTOS for developing a certain product. What this product
is and what it does is not relevant for my question.
I looked on the internet for a RTOS and ECOS is a possible chiose because it is
free. But what concerns me is that I find very different timing characteristics.
For example I found for Maximum Interrupt latency: 10us but also 100us. Is this
normal??
I will use a MPC555 from Motorola. Maybe anyone has some timing characteristics
about this processor with ECOS??
Thanks in advance.

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] ECOS Real-Time performance

[Gary Thomas]

On Thu, 2004-09-02 at 06:53, Meulendijks, J. wrote:
> Hi everybody,
> 
> I'm trying to choose a RTOS for developing a certain product. What this product
> is and what it does is not relevant for my question.
> I looked on the internet for a RTOS and ECOS is a possible chiose because it is
> free. But what concerns me is that I find very different timing characteristics.
> For example I found for Maximum Interrupt latency: 10us but also 100us. Is this
> normal??

Where did you find these numbers (10us / 100us)?  What platforms were
they for?

Normal for what?  That [particular] metric is very closely tied to a 
particular platform, not just a processor or architecture.  The factors
can can affect it include processor clock speed, memory design, 
peripheral mix, interrupt controller design, ...  If you want to use
this metric to compare operating systems, then you need to do it on
the *exact* target platform.  On the other hand, if you want to know
"can this system respond reliably within XXus of a realtime event?",
this number will give you an idea, but again only relative to the 
platforms on which the numbers were gathered.

Any OS that tells you "interrupt latency is 7.5us", without any 
explanation of the exact hardware platform, testing criteria, etc,
is trying to sell you the Brooklyn bridge.

With eCos, you can gather these data directly, not just "trust the
manufacturer".  Run 'tm_basic' with different criteria and see what
the numbers are on *your* target hardware.

> I will use a MPC555 from Motorola. Maybe anyone has some timing characteristics
> about this processor with ECOS??

Again, the numbers are suspect if they don't come from your platform.
I can line up a half dozen platforms with a particular processor, etc,
and come up with just as many results :-(  You may be able to get values
from a target similar to the one you want to run, but without measuring
on your platform, they are only *similar*.

Note: my point is that you need to look at more than simple numbers to
try and understand if a particular system is right for you.  
  * Does it have reasonable response times (not absolutes)?  
  * Does it support the correct programming model (e.g. eCos is a single
    process, multi-threaded model, with statically bound code)?
  * Are the features you need available (networking, file systems, ...)?
  * Is the target platform supported?  or easily ported to?

-- 
Gary Thomas <gary@mlbassoc.com>
MLB Associates


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

RE: [ECOS] ECOS Real-Time performance

[Meulendijks, J.]

  * Does it have reasonable response times (not absolutes)?  
  * Does it support the correct programming model (e.g. eCos is a single
    process, multi-threaded model, with statically bound code)?
  * Are the features you need available (networking, file systems, ...)?
  * Is the target platform supported?  or easily ported to?

All above point are covered by ECOS. And I know you are right about the timings
but I thought maybe someone tested it so they could give me their timings. But I
forgot to mention the platform.
How my systems looks like can be seen on
http://www.wuerz-elektronik.com/MPC555.html
But I will run 'tm_basic' to see the right timings. Thanks for your quick
reaction.

-----Original Message-----
From: Gary Thomas [mailto:gary@mlbassoc.com]
Sent: donderdag 2 september 2004 15:16
To: Meulendijks, J.
Cc: eCos Discussion
Subject: Re: [ECOS] ECOS Real-Time performance


On Thu, 2004-09-02 at 06:53, Meulendijks, J. wrote:
> Hi everybody,
> 
> I'm trying to choose a RTOS for developing a certain product. What this
product
> is and what it does is not relevant for my question.
> I looked on the internet for a RTOS and ECOS is a possible chiose because it
is
> free. But what concerns me is that I find very different timing
characteristics.
> For example I found for Maximum Interrupt latency: 10us but also 100us. Is
this
> normal??

Where did you find these numbers (10us / 100us)?  What platforms were
they for?

Normal for what?  That [particular] metric is very closely tied to a 
particular platform, not just a processor or architecture.  The factors
can can affect it include processor clock speed, memory design, 
peripheral mix, interrupt controller design, ...  If you want to use
this metric to compare operating systems, then you need to do it on
the *exact* target platform.  On the other hand, if you want to know
"can this system respond reliably within XXus of a realtime event?",
this number will give you an idea, but again only relative to the 
platforms on which the numbers were gathered.

Any OS that tells you "interrupt latency is 7.5us", without any 
explanation of the exact hardware platform, testing criteria, etc,
is trying to sell you the Brooklyn bridge.

With eCos, you can gather these data directly, not just "trust the
manufacturer".  Run 'tm_basic' with different criteria and see what
the numbers are on *your* target hardware.

> I will use a MPC555 from Motorola. Maybe anyone has some timing
characteristics
> about this processor with ECOS??

Again, the numbers are suspect if they don't come from your platform.
I can line up a half dozen platforms with a particular processor, etc,
and come up with just as many results :-(  You may be able to get values
from a target similar to the one you want to run, but without measuring
on your platform, they are only *similar*.

Note: my point is that you need to look at more than simple numbers to
try and understand if a particular system is right for you.  
  * Does it have reasonable response times (not absolutes)?  
  * Does it support the correct programming model (e.g. eCos is a single
    process, multi-threaded model, with statically bound code)?
  * Are the features you need available (networking, file systems, ...)?
  * Is the target platform supported?  or easily ported to?

-- 
Gary Thomas <gary@mlbassoc.com>
MLB Associates

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss