Huge performance issues

Huge performance issues

[Gwenn Boussard]

Hi,
We've got tens of standard PC's running an old Dos app' with Dosemu
(Trustix Linux distrib) since years.
Since a year or so, customers start complaining about performance on new
machines.
In fact browsing the text menu is slow that it's painfull !
A 'top' show that dosemu.bin use constantly 99% of CPU whatever I do.
On old machines it soars up to 40% at startup, then go down to 0% during
inactivity, and again up to few % during menu browsing.

The machines have various kernel and dosemu versions...

So I made some tests on two machines, a Fast and a Slow :
Put Fast dosemu binary on Slow
Put Slow kernel on Fast
Put Fast Dos image on Slow
Modified Slow dosemu.conf to look like Fast one
Verified final binaries were the same on Fast and Slow
And some other test I don't remember.

Whatever I do Slow stays slow, and Fast stays Fast.

There are tests I can't do because Slow is an off-site in production
machine, but I feel I tried all steps unsuccessfully.

Does anybody see another test I can do, any part that can go wrong ?
We don't use sound, nor mouse, just plain Dos emulation with IPX
support.

Regards,
Gwenn

Re: Huge performance issues

[Andrew Brooks]

Gwenn Boussard wrote:
> 
> Since a year or so, customers start complaining about performance on new
> machines.
> In fact browsing the text menu is slow that it's painfull !

I've seen something like that too!  

Using standard MS-DOS 6.2 and the MS-DOS text editor "edit"
takes seconds to respond to each key press.  dosemu itself is
generally normal speed, for example running commands at the
command prompt and starting "edit".  But pressing Page Down
for example takes seconds!

Does that sound like it's the same as your problem?

Andrew

Re: Huge performance issues

[Maximiliano Curia]

On Friday 17 June 2005 09:09, Gwenn Boussard wrote:
> In fact browsing the text menu is slow that it's painfull !
> A 'top' show that dosemu.bin use constantly 99% of CPU whatever I do.
> On old machines it soars up to 40% at startup, then go down to 0% during
> inactivity, and again up to few % during menu browsing.

Is the application a clipper program? Old clipper apps have a rather 
unfriendly to multitasking OS idle calls, if that's the case please try 
patching the app with: http://maxy.com.ar/~maxy/clipper

-- 
Saludos,
 /\/\ /\ >< `/

Re: Huge performance issues

[Stas Sergeev]

Hello.

Gwenn Boussard wrote:
> In fact browsing the text menu is slow that it's painfull !
> A 'top' show that dosemu.bin use constantly 99% of CPU whatever I do.
Looks like you forgot to specify
the dosemu version you are using/tried.
Without this, all the other information
is completely irrelevant.

Re: Huge performance issues

[Stas Sergeev]

Hello.

Andrew Brooks wrote:
> generally normal speed, for example running commands at the
> command prompt and starting "edit".  But pressing Page Down
> for example takes seconds!
> Does that sound like it's the same as your problem?
This sounds like you forgot to type
speed 0
or setting
$_hogthreshold=(0)
before starting "edit".

Re: Huge performance issues

[Hufnus]

On Fri, 17 Jun 2005 14:09:53 +0200
Gwenn Boussard <gwenn@talc.fr> wrote:

> A 'top' show that dosemu.bin use constantly 99% of CPU whatever I
> do. On old machines it soars up to 40% at startup, then go down to
> 0% during

Verified, regardless of $_hogthreshold setting, top shows about
89 - 93% utilization all the time with 1.3.2.0

With 1.1.4, it idles at 8.6% and when it has I/O in progress or
pending it goes as high as 72%

I am switching back, had a lot of incompatibility issues since 1.2.0!
Old dosemu handles the screen faster and better and, for example, it
does not screw the screen during boot, when you use a dynamic
config.sys editor like Ziff's config.ctl and general response is
snappier.

TonyB

--
  __      __  _     I N C.               http://www.sysdev.org
/ __|\\// __||  \  __   __          /         tonyb@sysdev.org
\__ \ \/\__ \||)|/ O_)\/ /        \/  System Tools / Utilities
|___/ || ___/|_ /\___|\_/        WIntel / Linux Device Drivers

Re: Huge performance issues

[Stas Sergeev]

Hello.

Hufnus wrote:
> Verified, regardless of $_hogthreshold setting, top shows about
> 89 - 93% utilization all the time with 1.3.2.0
It will almost certainly drop down
if you change in
src/base/misc/ioctl.c:152
reset_idle(10) to something like
reset_idle(2) or reset_idle(0).
I think 2 should quite work.

> I am switching back, had a lot of
> incompatibility issues since 1.2.0!
While finding the most suitable
version is the proper way of solving
some problems, screaming "I am
switching back!" in the public ML,
instead of to report the bug as most
users do, is probably not the best
way of getting any help.
And since so far what you called an
"incompatibility" was only the fact
that the old dosemu was ignoring some
wrong configs, while the new one
terminates - there is probably not
a big deal if you use only the stable
versions of any software.

Re: Huge performance issues

[Gwenn Boussard]

> if you change in
> src/base/misc/ioctl.c:152
> reset_idle(10) to something like
> reset_idle(2) or reset_idle(0).
> I think 2 should quite work.
It works !
I haven't time to do extensive testing, so I don't know what the problem
came from, but compiling 1.3.2 with reset_idle(2) just works.
For your information, several versions of Dosemu have been reported
having troubles.
The problem is that I haven't access to all of the boxes, the only one I
can look at was a 1.1.99.1

I will stay with this 1.3.2 until troubles are reported.

Thanks for all,
Gwenn

Re: Huge performance issues

[Stas Sergeev]

Hello.

Gwenn Boussard wrote:
>> if you change in
>> src/base/misc/ioctl.c:152
>> reset_idle(10) to something like
>> reset_idle(2) or reset_idle(0).
>> I think 2 should quite work.
> It works !
Thanks for testing, though I am
surprised it works *for you*.
You said you tried different
versions, and I am pretty sure
this line exists only in 1.3.2
(that's why not apecifying the
versions was a bad decision).
So it can be something else after
all.

> I will stay with this 1.3.2 until troubles are reported.
Since you all "scared" me, I put
reset_idle(0) in CVS, not even 2.
So if you can test also with 0 -
that would be nice (and try the
different $_hogthreshold values
too, if something goes wrong),
but also double-check that the
reset_idle(10) is exactly an issue
for you, since it wasn't there
before and so I am surprised how
it could affect the other versions.

Re: Huge performance issues

[Gwenn Boussard]

> Thanks for testing, though I am
> surprised it works *for you*.
> You said you tried different
> versions, and I am pretty sure
> this line exists only in 1.3.2
> (that's why not apecifying the
> versions was a bad decision).
> So it can be something else after
> all.
Other tested versions were stock version without any tweak.
I must admit I did test in a hurry and can't exclude mistake from myself
that would explain why none of tested versions worked.
The only things I'm really sure is 1.1.99.1 didn't work correctly and
that 1.3.2 with reset_idle(2) does.

> > I will stay with this 1.3.2 until troubles are reported.
> Since you all "scared" me, I put
> reset_idle(0) in CVS, not even 2.
> So if you can test also with 0 -
> that would be nice (and try the
> different $_hogthreshold values
> too, if something goes wrong),
> but also double-check that the
> reset_idle(10) is exactly an issue
> for you, since it wasn't there
> before and so I am surprised how
> it could affect the other versions.
I'm afraid I can't do that. The Box is in production and I won't touch
it unless there's a very good reason to.
And I haven't time to build a test box.

I have appreciated your help. Thanks,
Gwenn