bash performance not so gooood

bash performance not so gooood

[Khosla, Deepak]

Hi,
Just some comparison numbers on a large script I ran on Windows NT 4.0 
SP3 using some different packages:

Gnuwin bash	- 2 mins 20 secs
MKS ksh	- 2 mins 0 secs
ATT UWIN ksh -  0 mins 20 secs

Any opinions as to why?
Also, the CPU utilization was extremely high and memory usage by the 
sh.exe (sometimes several probably where there are pipes in the 
script) was around 5-6Mb each! Not a pretty sight.

I was running on dual 200MHz Pentium with 128MB ram.
Gnuwin32 B18 with the coolview patch (same performance without 
patch).

Any ideas since at this performance, I would have to buy the UWIN 
package and I would prefer not doing it....

Regards

Deepak Khosla
281-514-9234
Deepak.Khosla@compaq.com


-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request@cygnus.com" with one line of text: "help".

Re: bash performance not so gooood

[Marty Leisner]

In message <c=US%a=_%p=COMPAQ%l=EXCHOU-CA0201-970917182110Z-7831@exchou-conn02.
im.hou.compaq.com>,   you write:
>Hi,
>Just some comparison numbers on a large script I ran on Windows NT 4.0 
>SP3 using some different packages:
>
>Gnuwin bash	- 2 mins 20 secs
>MKS ksh	- 2 mins 0 secs
>ATT UWIN ksh -  0 mins 20 secs
>
>Any opinions as to why?
>Also, the CPU utilization was extremely high and memory usage by the 
>sh.exe (sometimes several probably where there are pipes in the 
>script) was around 5-6Mb each! Not a pretty sight.
>

Can you also run on djgpp bash?
And a unix host (linux preferably?)

marty

-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request@cygnus.com" with one line of text: "help".

Re: bash performance not so gooood

[Geoffrey Noer]

Khosla, Deepak wrote:
> 
> Hi,
> Just some comparison numbers on a large script I ran on Windows NT 4.0 
> SP3 using some different packages:
> 
> Gnuwin bash	- 2 mins 20 secs
> MKS ksh	- 2 mins 0 secs
> ATT UWIN ksh -  0 mins 20 secs
> 
> Any opinions as to why?

I'm not surprised.

Very little time has been spent optimizing cygwin.dll or bash.
Bash currently uses fork() instead of spawn() which costs it
quite a bit of time.  The Cygwin32 layer also needs performance
analysis.  Until that happens, it will be slower than it needs to
be.

> Any ideas since at this performance, I would have to buy the UWIN 
> package and I would prefer not doing it....

You could either buy UWIN or improve cygwin.dll and/or bash and send
us patches.  :-)

-- 
Geoffrey Noer
noer@cygnus.com
-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request@cygnus.com" with one line of text: "help".

Re: bash performance not so gooood

[Alex]

On Wed, 17 Sep 1997, Geoffrey Noer wrote:

> Very little time has been spent optimizing cygwin.dll or bash.
> Bash currently uses fork() instead of spawn() which costs it
> quite a bit of time.  The Cygwin32 layer also needs performance
> analysis.  Until that happens, it will be slower than it needs to
> be.
> 
> > Any ideas since at this performance, I would have to buy the UWIN 
> > package and I would prefer not doing it....
> 
> You could either buy UWIN or improve cygwin.dll and/or bash and send
> us patches.  :-)

Has anyone considered running ash and not bash as sh, which unless I'm
dead wrong, might speed some configure scripts up as it's much smaller.
Also, how exactly does spawn work (It's been a while
since I've used Win95 and FreeBSD doesn't have a spawn command) vs
fork/exec?

- alex

-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request@cygnus.com" with one line of text: "help".

RE: bash performance not so gooood

[Alex]

On Thu, 18 Sep 1997, Sergey Okhapkin wrote:

> dahms@ifk20.mach.uni-karlsruhe.de wrote:
> > 
> > I'd like to know the results for OpenNT 2.0, too.
> > However, OpenNT replaces the MS posix.dll, and so does, I think, UWIN...
> 
> UWIN is not a posix subsystem like OpenNT. UWIN runs on top of Win32 API like cygwin.
> 
> > But without doubt, Linux on the same box would win hands down 8-)
> 
> Shell scripts runs much faster on linux, but cpu eating tasks (like compilation) requires almost the same time.

Actually, no.  FreeBSD seems much faster on two different machines than 95
was at compiling.  Using -pipe on fbsd made things even faster.
 
- alex

-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request@cygnus.com" with one line of text: "help".

RE: bash performance not so gooood

[Sergey Okhapkin]

dahms@ifk20.mach.uni-karlsruhe.de wrote:
> 
> I'd like to know the results for OpenNT 2.0, too.
> However, OpenNT replaces the MS posix.dll, and so does, I think, UWIN...

UWIN is not a posix subsystem like OpenNT. UWIN runs on top of Win32 API like cygwin.

> But without doubt, Linux on the same box would win hands down 8-)

Shell scripts runs much faster on linux, but cpu eating tasks (like compilation) requires almost the same time.

-- 
Sergey Okhapkin, http://www.lexa.ru/sos
Moscow, Russia
Looking for a job

-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request@cygnus.com" with one line of text: "help".

RE: bash performance not so gooood

[Sergey Okhapkin]

Khosla, Deepak wrote:
> Hi,
> Just some comparison numbers on a large script I ran on Windows NT 4.0
> SP3 using some different packages:
>
> Gnuwin bash	- 2 mins 20 secs
> MKS ksh	- 2 mins 0 secs
> ATT UWIN ksh -  0 mins 20 secs
>
> Any opinions as to why?

CreateProcess() syscall is very slow on Windows NT. All the UWIN stuff is 
optimized to make spawn() calls instead of usual fork/exec pairs (both fork 
and exec makes CreateProcess()). Moreover, path/symlinks handling in very 
complex and ugly in cygwin and should to be rewritten from the scratch...
Yesterday I've made the performance comparison running configure script for 
bash 2.01 with cygwin and OpenNT 2.0+gcc. The result was 4 mins 43 secs on 
OpenNT and 11 mins 20 secs on cygwin :-(

--
Sergey Okhapkin, http://www.lexa.ru/sos
Moscow, Russia
Looking for a job

-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request@cygnus.com" with one line of text: "help".

RE: bash performance not so gooood

[dahms]

Hi, you wrote:

: Gnuwin bash	- 2 mins 20 secs
: MKS ksh	- 2 mins 0 secs
: ATT UWIN ksh -  0 mins 20 secs

I'd like to know the results for OpenNT 2.0, too.
However, OpenNT replaces the MS posix.dll, and so does, I think, UWIN...
But without doubt, Linux on the same box would win hands down 8-)


Bye, Heribert (dahms@ifk20.mach.uni-karlsruhe.de)
-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request@cygnus.com" with one line of text: "help".