Ada for OS/2 (was Re: Linux faster than OS/2...)

Ada for OS/2 (was Re: Linux faster than OS/2...)

[system]

I am crossposting this to comp.lang.ada, comp.os.os2.setup.misc,
comp.os.os2.programmer.misc, comp.os.os2.programmer.oop

erdmann@ibm.net writes:
>"@ribs.ssesco.com (" William Nau") writes:
>>: I seems i should use Linux when it comes done to processor intensive 
>>: jobs like numerical jobs. 

>>I think the conclusion is drawn too quickly.

>Just to clarify, how i concluded this figure. I did a compilation of
>a package of 5 Ada 95 modules and compiled this on both Systems. The 
>difference of the elapse time was 25 %.

I would not necessily expect OS/2's compiler to have the same effeciency
as Linux's.  Assuming the hardware is identical the other question I have 
(not having read the original post) is whether you were running a GUI on 
Linux?

=====================================================================

This does bring up another question, GNAT provides a free ADA compilier
for OS/2 (among other OSes).   I am given to believe that it was
originally written with gov't funding but currently is only upgraded
because one of the primary programmers likes OS/2.

One way to improve the attention given to the OS/2 port and assure its
continued existance would be for "somebody" to purchase a maintenance
contract with ACT (the company that writes GNAT).  I don't know
the cost but I am fairly certain that this would be rather a bit
much for an individual but if we could pool 20-100 people who are
willing to chip in a not small amount on an ongoing basis
(I am _GUESSING_ at $50/year or so depending on lots of things)...

Robert

Morphis@physics.niu.edu
Real Men change diapers

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Robert Dewar]

Ralph asks

<<What about the ext2-os2 file system drivers. Does the use of a Linux
style
file system boost the performance ?
>>

In the only experiment we did on this, using the linux style file system
substantially slowed down compilation speed on OS/2.

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Robert Dewar]

Robert said

<<I would not necessily expect OS/2's compiler to have the same effeciency
as Linux's.  Assuming the hardware is identical the other question I have
(not having read the original post) is whether you were running a GUI on
Linux?>>

  We find some things compile faster in OS/2, and others compile faster
  in Linux. Be careful that memory and disk support are similar. One
  factor that hurts OS/2 noticebly is the pathetic limit of 2 megs on
  the HPFS cache size. In some situations, using FAT on OS/2 will speed
  up compilations due to the bigger cache size.

<<This does bring up another question, GNAT provides a free ADA compilier
for OS/2 (among other OSes).   I am given to believe that it was
originally written with gov't funding but currently is only upgraded
because one of the primary programmers likes OS/2.>>

  GNAT is free software in the GNU sense, it is a front end for the
  GCC compiler, which has been adapted to handle Ada 95. The port is
  maintained by Ada Core Technologies, and it is maintained because
  we have commercial customers who provide sufficient support to make
  this continued maintenance supportable for us. It is also true that
  one of the primary programmers likes OS/2 (me!)

<<One way to improve the attention given to the OS/2 port and assure its
continued existance would be for "somebody" to purchase a maintenance
contract with ACT (the company that writes GNAT).  I don't know
the cost but I am fairly certain that this would be rather a bit
much for an individual but if we could pool 20-100 people who are
willing to chip in a not small amount on an ongoing basis
(I am _GUESSING_ at $50/year or so depending on lots of things)...>>

  Well our maintenance contracts are based on the total number of
  programmers using Ada, so this model is not applicable. But in any
  case, fear not! The OS/2 port is fully maintained. You can obtain
  the public version of the compiler from many FTP sites, including
  cs.nyu.edu. If you are interested in commercial support for GNAT
  on OS/2 or any other platform, send email to info@gnat.com for
  an electronic copy of our brochure, or visit www.gnat.com.

  P.S. If you really think $5000/year would support this work, you are
  somewhat off in your calculation. That is about 2-3 weeks of work in
  a year (we are a commercial company, not hobbyists fiddling in our
  spare time). The cost of maintaining the OS/2 port is very much higher
  than that. However, as I say, no need to worry, it looks like, at least
  for now, there is (just) enough commercial interest to keep the OS/2
  port viable (it helps of course that I like OS/2 :-)

Robert Dewar
Ada Core Technologies

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Robert Dewar]

Robert said

<<I would not necessily expect OS/2's compiler to have the same effeciency
as Linux's.  Assuming the hardware is identical the other question I have
(not having read the original post) is whether you were running a GUI on
Linux?
>>

One clarification here. The compilers running on OS/2 and Linux are
virtually identical code. So any differences you see in performance are
not attributable to the compiler itself, but to OS and hardware 
considerations. As I noted in my previous post, my guess is that, if
you are running on identical hardware, the most likely difference comes
from the small HPFS cache size in OS/2.

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[system]

dewar@merv.cs.nyu.edu (Robert Dewar) writes:

>Robert said

><<I would not necessily expect OS/2's compiler to have the same effeciency
>as Linux's.  Assuming the hardware is identical the other question I have
>(not having read the original post) is whether you were running a GUI on
>Linux?>>

>  We find some things compile faster in OS/2, and others compile faster
>  in Linux. Be careful that memory and disk support are similar. One
>  factor that hurts OS/2 noticebly is the pathetic limit of 2 megs on
>  the HPFS cache size. 

<heh> Gee, where have I heard that complaint before?

><<This does bring up another question, GNAT provides a free ADA compilier
>for OS/2 (among other OSes).   I am given to believe that it was
>originally written with gov't funding but currently is only upgraded
>because one of the primary programmers likes OS/2.>>
>
>  GNAT is free software in the GNU sense, it is a front end for the
>  GCC compiler, which has been adapted to handle Ada 95. The port is
>  maintained by Ada Core Technologies, and it is maintained because
>  we have commercial customers who provide sufficient support to make
>  this continued maintenance supportable for us. 

Ah, my mistake :(

>  It is also true that one of the primary programmers likes OS/2 (me!)

thought so, but didn't care to put my foot in my mouth any more than
necessary.

>  Well our maintenance contracts are based on the total number of
>  programmers using Ada, so this model is not applicable. 

I had assumed that something would have to be worked out, but
thankfully we don't have to worry about it.

>  The cost of maintaining the OS/2 port is very much higher
>  than [$5000/year]

Interesting.

Thanks for the reply,

Robert
Morphis@physics.niu.edu
Real Men change diapers

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Ralph Paul]

Robert Dewar wrote:
> 
> One clarification here. The compilers running on OS/2 and Linux are
> virtually identical code. So any differences you see in performance are
> not attributable to the compiler itself, but to OS and hardware
> considerations. As I noted in my previous post, my guess is that, if
> you are running on identical hardware, the most likely difference comes
> from the small HPFS cache size in OS/2.

What about the ext2-os2 file system drivers. Does the use of a Linux
style
file system boost the performance ?


CU/2,

Ralph Paul

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Jerry van Dijk]

In <33BABA02.2762@ibm.net>, Ralph Paul <repaul@ibm.net> writes:

>What about the ext2-os2 file system drivers. Does the use of a Linux
>style
>file system boost the performance ?

If you have sufficient RAM (64Mb or more I would guess) setting:

        gccload=10
        gccopt=-pipe

does give quite a boost.

-- Jerry van Dijk | email: jdijk@acm.org
-- Consultant     |       TEAM ADA
-- Ordina Finance |    Leiden, Holland

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Geert Bosch]

In comp.lang.ada system@niuhep.physics.niu.edu wrote:
 ``I would not necessily expect OS/2's compiler to have the same effeciency
   as Linux's.  Assuming the hardware is identical the other question I have 
   (not having read the original post) is whether you were running a GUI on 
   Linux?''

Please use emxload -e -gnat before doing any performance tests. This
keeps the GNAT binaries in VM (real memory or swap space) instead
of reloading and relocating them every time they are needed. This
makes a huge difference in speed, esp.  with small files. 
Because filesystem caching in Linux is better, this is less of an issue
with Linux. 

Also keep in mind that you are mostly testing operating systems and I/O
speed. The actual code that will be executed is almost the same, so given
enough memory and fast enough I/O the performance will be not much
different on Linux as on OS/2.

Regards,
   Geert

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Matthieu Willm]

What max. cache size did you use with ext2-os2 ? The default is 256 Kb
and is small. But with larger cache sizes (4 Mb on my 40 Mb system)
compilation speed of ext2-os2 itself is quite fast.

Robert Dewar wrote:
> 
> Ralph asks
> 
> <<What about the ext2-os2 file system drivers. Does the use of a Linux
> style
> file system boost the performance ?
> >>
> 
> In the only experiment we did on this, using the linux style file system
> substantially slowed down compilation speed on OS/2.

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Robert Dewar]

Haug said

<<and set "SET GCCOPT=-pipe" in your config.sys to prevent
gnat from using tempfiles. It uses pipes instead.
>>

I tried this, it made absolutely zero difference in speed. Are you just
guessing that this should speed up GNAT, or is that your actual experience.

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Geert Bosch]

[Some newsgroups removed]

In comp.lang.ada erdmann@ibm.net wrote:
   I have re done my tests again based on the information in mind
   i did receive so far. The results for a 5 packages OS/2 PM
   program are:

           OS/2 : 107 sec   
           Linux:  95 sec

What you should look at is CPU load during compile which should be
100%. A lower figure indicates that the system is waiting for I/O,
which can be caused by a diskcache that is too small, disabled
lazy-write caching, or swapping due to insufficient memory. 

When CPU-load is straight 100% you know you can't do much better.
In that case the difference of 10% seems high to me for compilation.
(Binding and linking is another story, the binder is I/O bound and
the linker is system-dependent.)

Regards,
   Geert

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[erdmann]

In <5peljs$i2p$1@gonzo.sun3.iaf.nl>, Geert Bosch <geert@gonzo.sun3.iaf.nl> writes:
>In comp.lang.ada system@niuhep.physics.niu.edu wrote:
> ``I would not necessily expect OS/2's compiler to have the same effeciency
>   as Linux's.  Assuming the hardware is identical the other question I have 
>   (not having read the original post) is whether you were running a GUI on 
>   Linux?''
>
>Please use emxload -e -gnat before doing any performance tests. This
>keeps the GNAT binaries in VM (real memory or swap space) instead
>of reloading and relocating them every time they are needed. This
>makes a huge difference in speed, esp.  with small files. 
>Because filesystem caching in Linux is better, this is less of an issue
>with Linux. 
>
>Also keep in mind that you are mostly testing operating systems and I/O
>speed. The actual code that will be executed is almost the same, so given
>enough memory and fast enough I/O the performance will be not much
>different on Linux as on OS/2.
>
I have re done my tests again based on the information in mind i did receive
so far. The results for a 5 packages OS/2 PM program are:

        OS/2 : 107 sec   
        Linux:  95 sec

In order to get this result i did close all windows of the WPS displaying
the filesystem. Now the difference in performance is about 10 percent. But
this seems to be the best!

Michael

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Haug Buerger]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 908 bytes --]


On 3 Jul 1997 00:44:12 +0200, Geert Bosch <geert@gonzo.sun3.iaf.nl> wrote:
>In comp.lang.ada system@niuhep.physics.niu.edu wrote:
> ``I would not necessily expect OS/2's compiler to have the same effeciency
>   as Linux's.  Assuming the hardware is identical the other question I have
>   (not having read the original post) is whether you were running a GUI on
>   Linux?''
>
>Please use emxload -e -gnat before doing any performance tests. This
>keeps the GNAT binaries in VM (real memory or swap space) instead
>of reloading and relocating them every time they are needed. This
>makes a huge difference in speed, esp.  with small files.
>Because filesystem caching in Linux is better, this is less of an issue
>with Linux.

and set "SET GCCOPT=-pipe" in your config.sys to prevent
gnat from using tempfiles. It uses pipes instead.

Haug

--

Save a tree use email to haug@zesi.ruhr.de (Haug B�rger)

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Michael Erdmann]

Robert Dewar wrote:
> 
> Robert said
> 
> <<I would not necessily expect OS/2's compiler to have the same effeciency
> as Linux's.  Assuming the hardware is identical the other question I have
> (not having read the original post) is whether you were running a GUI on
> Linux?
> >>
> 
> One clarification here. The compilers running on OS/2 and Linux are
> virtually identical code. So any differences you see in performance are
> not attributable to the compiler itself, but to OS and hardware
> considerations. As I noted in my previous post, my guess is that, if
> you are running on identical hardware, the most likely difference comes
> from the small HPFS cache size in OS/2.
I doubt this, because the cache of my Linux installtion is about 2 MB.
Earlier in the discussion some body mentioned the fact, that the emx
dll are ported from unix to OS/2 in a inefficient way. Could this
the probem ?

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Robert Dewar]

Michael said

<<I have done this today. It seems now to work a bit faster.>>

That's not very convincing :-)
Did you actually measure the difference.
I tried the -pipe and it made no difference at all to the 
compilation speed of several examples I tried.

But a lot depends on your setup. In my setup, I have 80 megs and never
swap, so in fact disk IO is completely overlapped (I tried an interesting
experiment which was to put

sources, objects, temp files, ali files

for the library all in a RAM disk, and compile the library. It took, 
within measurable accuracy EXACTLY the same time as using a disk with
the normal HPFS cache.

That's why I don't think -pipe will help. If you have a decent amount
of RAM and a 2 meg cache, then the temporary files get written to the
cache, and read from the cache. Sure, they get written to disk as well,
but these are lazy writes from the cache which can be completely
overlapped with computation.

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Michael Erdmann]

Robert Dewar wrote:
> 
> Michael said
> 
> <<I have done this today. It seems now to work a bit faster.>>
> 
> That's not very convincing :-)
> Did you actually measure the difference.
> I tried the -pipe and it made no difference at all to the
> compilation speed of several examples I tried.
I have repeated the measurement. The difference is not measurable,
what happend was more simple. I did increase my ram cache from
1.5M to 2M and this caused most of the enhancement in the speed.

> 
> But a lot depends on your setup. In my setup, I have 80 megs and never
> swap, so in fact disk IO is completely overlapped (I tried an interesting
> experiment which was to put
> 
> sources, objects, temp files, ali files
> 
> for the library all in a RAM disk, and compile the library. It took,
> within measurable accuracy EXACTLY the same time as using a disk with
> the normal HPFS cache.
> 
I dont have 80 MB, but i allready thought about installing a ram disk
for the same purpose.

> That's why I don't think -pipe will help. If you have a decent amount
> of RAM and a 2 meg cache, then the temporary files get written to the
> cache, and read from the cache. Sure, they get written to disk as well,
> but these are lazy writes from the cache which can be completely
> overlapped with computation.
I agree on this and i am not going to install any ramsdisk.

By the way i have redone some measurement based upon a software package 
called rudstone. I did take the sources from the 1995 Ada CDROM from
Walnut Creek and done the same measurements:

		Compilation+Build	Execution 
	OS/2            150 sec		   97 
	Linux		135 sec		  114

Compiling is slower in OS/2 but the execution time for the benchmark
is smaler, meaning better performance. 
What i understood so far, i have to accept some performance degrade
for the compilation in favor of the OS/2 features, but the performance 
of programms is not automaticaly less good then with Linux. 
As a result i decided to stay with OS/2 instead of switching to Linux.

Michael

Re: Ada for OS/2 (was Re: Linux faster than OS/2...)

[Michael Erdmann]

Geert Bosch wrote:
> 
> [Some newsgroups removed]
> 
> In comp.lang.ada erdmann@ibm.net wrote:
>    I have re done my tests again based on the information in mind
>    i did receive so far. The results for a 5 packages OS/2 PM
>    program are:
> 
>            OS/2 : 107 sec
>            Linux:  95 sec
> 
> What you should look at is CPU load during compile which should be
> 100%. A lower figure indicates that the system is waiting for I/O,
> which can be caused by a diskcache that is too small, disabled
> lazy-write caching, or swapping due to insufficient memory.
> 
> When CPU-load is straight 100% you know you can't do much better.
> In that case the difference of 10% seems high to me for compilation.
> (Binding and linking is another story, the binder is I/O bound and
> the linker is system-dependent.)

The CPU is 100 % under load!

Michael

Re: Ada for OS/2

[Geert Bosch]

A few weeks ago, Haug said
   <<and set "SET GCCOPT=-pipe" in your config.sys to prevent
   gnat from using tempfiles. It uses pipes instead.>>

Robert Dewar replied:
   <<I tried this, it made absolutely zero difference in speed.
   Are you just guessing that this should speed up GNAT, or is that
   your actual experience.>>

One last note: not only does using SET GCCOPT=-pipe not have any
effect on compilation speed, it does introduce some problems. When
GNAT generates informal messages, these messages get passed to the
assembler input with an error as result. The reason is that these
informal messages are written to standard out, just like the assembly
code when using pipes.

This happens when a compilation generates no code (compiling a
generic for example) and when using -gnatv. So I would recommend
not to use GCCOPT=-pipe with GNAT.

Regards,
   Geert

Re: Ada for OS/2

[Tarjei T. Jensen]

Geert Bosch wrote:
> One last note: not only does using SET GCCOPT=-pipe not have any
> effect on compilation speed, it does introduce some problems. When
> GNAT generates informal messages, these messages get passed to the
> assembler input with an error as result. The reason is that these
> informal messages are written to standard out, just like the assembly
> code when using pipes.

This is rather disturbing as I would expect messages from the compiler
to be written to stderr. Especially when stdout is other use.

I would think that the handling of messages in the compiler needs some
adjustments with regards to where they go.

Greetings,

-- 
// Tarjei T. Jensen 
//    tarjei@online.no || voice +47 51 62 85 58
//   Support you local rescue centre: GET LOST!

Re: Ada for OS/2

[Robert Dewar]

Tarjei said

<<This is rather disturbing as I would expect messages from the compiler
to be written to stderr. Especially when stdout is other use.

I would think that the handling of messages in the compiler needs some
adjustments with regards to where they go.>>


Read the documentation. Long form messages including the source go to
stdout (since nothing else goes to stdout from the compiler, this makes 
perfectly good sense, also the compiler listing goes to stdout).
Brief form error messages go to stderr.

The use of the pipe option is completely useless in any case, as I
pointed out a while ago, so the fact that it does not work is hardly
significant!