Project Home
Project Home
Source Code
Source Code
Documents
Documents
Wiki
Wiki
Discussion Forums
Discussions
Project Information
Project Info
Forum Topic - Windows much slower then self-hosted: (11 Items)
   
Windows much slower then self-hosted  
We have our custom makefile. From the root of the procect invoking make clean takes 6 seconds on QNX, 15 on XP (much 
slower computer ) ,  45seconds on  Windows on Vista64 which is the fastest of them all has has the same model of HD as 
the QNX machine.

Any idea what is going on . I've disable Windows Defender and UAC but no difference.  CPU is basically idle while 
performing the clean up.
Re: Windows much slower then self-hosted  
Heh - probably faster to compile in a VM machine? ;-)

Mario Charest wrote:
> We have our custom makefile. From the root of the procect invoking make clean takes 6 seconds on QNX, 15 on XP (much 
slower computer ) ,  45seconds on  Windows on Vista64 which is the fastest of them all has has the same model of HD as 
the QNX machine.
> 
> Any idea what is going on . I've disable Windows Defender and UAC but no difference.  CPU is basically idle while 
performing the clean up.
> 
> _______________________________________________
> General
> http://community.qnx.com/sf/go/post18332
> 

-- 
cburgess@qnx.com
Re: Windows much slower then self-hosted  
Mario Charest wrote:
>  We have our custom makefile. From the root of the procect invoking
>  make clean takes 6 seconds on QNX, 15 on XP (much slower computer ) ,
>  45seconds on  Windows on Vista64 which is the fastest of them all has
>  has the same model of HD as the QNX machine.
>
>  Any idea what is going on . I've disable Windows Defender and UAC but
>  no difference.  CPU is basically idle while performing the clean up.

If you mean what is going on with the Vista machine, Vista's IO subsystem is
perceptively an order of magnitude slower than XP. 

If you go and read into the details from various MS developers, 
apparently the actual
throughput is higher, but in Vista it was determined that you would 
actually 'see' the real
progress vs when stuff was 'cached well enough'.

My experience is that even with Vista SP1, the perceptive speed is still 
way off of XP.
You can do lots of UI tuning to get small gains, but I rolled back to XP 
to get the build
speed.

Thomas
RE: Windows much slower then self-hosted  
> 
> If you mean what is going on with the Vista machine, Vista's IO
> subsystem is
> perceptively an order of magnitude slower than XP.
> 
> If you go and read into the details from various MS developers,
> apparently the actual
> throughput is higher, but in Vista it was determined that you would
> actually 'see' the real
> progress vs when stuff was 'cached well enough'.
> 
> My experience is that even with Vista SP1, the perceptive speed is
> still way off of XP.


> You can do lots of UI tuning to get small gains,

UI tuning?  I don't see what that has to do with a full console based operation ( I ran make via the command prompt )

> but I rolled back to XP to get the build speed.

For some reason I'm not ready to point the finger at Vista.  There a few things I would like to figure out first. For 
example, even though I'm using -j4 I've spotted up to 14 qcc.exe and cc1plus.exe in memory at one time (some seems to be
 zombie like). The tool i'm using, TaskInfo, doesn't show any open files for cc1plus.exe, even the one that aren't 
zombie. Could it be because they go through the cygwin thingy magiger.

Could this be an issue with licensing?  I mean since CPU is not being used it has to be waiting for something, could be 
the HD.  I'm trying to find that out.

> 
> Thomas
> 
> 
> _______________________________________________
> General
> http://community.qnx.com/sf/go/post18341
> 
Re: Windows much slower then self-hosted  
Mario Charest wrote:
> > If you mean what is going on with the Vista machine, Vista's IO
> > subsystem is perceptively an order of magnitude slower than XP.
> >
> > If you go and read into the details from various MS developers,
> > apparently the actual throughput is higher, but in Vista it was
> > determined that you would actually 'see' the real progress vs when
> > stuff was 'cached well enough'.
> >
> > My experience is that even with Vista SP1, the perceptive speed is
> > still way off of XP.
>
> > You can do lots of UI tuning to get small gains,
>
>  UI tuning?  I don't see what that has to do with a full console based
>  operation ( I ran make via the command prompt )

In my tests I was running console builds but they were running within the
"graphical interface" of Windows so the text scrolling and buffer 
maintenance
within the rich Aero user interface contributed to the overhead.   Of course
I could minimize the build console window, but where is the fun in that.

> > but I rolled back to XP to get the build speed.
>
>  For some reason I'm not ready to point the finger at Vista.  There a
>  few things I would like to figure out first. For example, even though
>  I'm using -j4 I've spotted up to 14 qcc.exe and cc1plus.exe in memory
>  at one time (some seems to be zombie like). The tool i'm using,
>  TaskInfo, doesn't show any open files for cc1plus.exe, even the one
>  that aren't zombie. Could it be because they go through the cygwin
>  thingy magiger.
>
>  Could this be an issue with licensing?  I mean since CPU is not being
>  used it has to be waiting for something, could be the HD.  I'm trying
>  to find that out.

I've used perfmon (Control Panel > Administrative Tools) to watch what
is going on with different aspects of the system.   I'd be interested to 
hear
what you find since my tests have been on a clean XP vs Vista system with
the exact same HW and same QNX SW and it was "go get a coffee" vs
"go eat lunch" kind of time differences in what I was seeing on my projects.

Thomas
RE: Windows much slower then self-hosted  
> Mario Charest wrote:
> > > If you mean what is going on with the Vista machine, Vista's IO
> > > subsystem is perceptively an order of magnitude slower than XP.
> > >
> > > If you go and read into the details from various MS developers,
> > > apparently the actual throughput is higher, but in Vista it was
> > > determined that you would actually 'see' the real progress vs when
> > > stuff was 'cached well enough'.
> > >
> > > My experience is that even with Vista SP1, the perceptive speed is
> > > still way off of XP.
> >
> > > You can do lots of UI tuning to get small gains,
> >
> >  UI tuning?  I don't see what that has to do with a full console
> based
> >  operation ( I ran make via the command prompt )
> 
> In my tests I was running console builds but they were running within
> the
> "graphical interface" of Windows so the text scrolling and buffer
> maintenance
> within the rich Aero user interface contributed to the overhead.  

The make session I run last 45 seconds on Vista, and it displays 90 lines ;-)


 Of
> course
> I could minimize the build console window, but where is the fun in
> that.
> 
> > > but I rolled back to XP to get the build speed.
> >
> >  For some reason I'm not ready to point the finger at Vista.  There a
> >  few things I would like to figure out first. For example, even
> though
> >  I'm using -j4 I've spotted up to 14 qcc.exe and cc1plus.exe in
> memory
> >  at one time (some seems to be zombie like). The tool i'm using,
> >  TaskInfo, doesn't show any open files for cc1plus.exe, even the one
> >  that aren't zombie. Could it be because they go through the cygwin
> >  thingy magiger.
> >
> >  Could this be an issue with licensing?  I mean since CPU is not
> being
> >  used it has to be waiting for something, could be the HD.  I'm
> trying
> >  to find that out.
> 
> I've used perfmon (Control Panel > Administrative Tools) to watch what
> is going on with different aspects of the system.   I'd be interested
> to hear what you find since my tests have been on a clean XP vs Vista system
> with the exact same HW and same QNX SW and it was "go get a coffee" vs
> "go eat lunch" kind of time differences in what I was seeing on my
> projects.

I haven`t tried XP on that machine.  So I can`t compare but my machine is a QuadCore running a 3.4 Gig (45 seconds) an 
on a co-worker XP laptop Dual core running a 1.8Gig it takes 15 seconds.

I ran HDTune to check the performance of the HD.  The average is 44Meg but very often it dips below 10Megs, reaching as 
low as 2Mbyte/sec.  Maybe I got a hardware/config problem with the machine.

> 
> Thomas
> 
> 
> _______________________________________________
> General
> http://community.qnx.com/sf/go/post18372
> 
Re: RE: Windows much slower then self-hosted  
Is there any update for this problem?  I am also seeing this issue while building on Windows, in this case it is Windows
 7 Enterprise 64-bit.  I did some digging into the problem and have narrowed the problem as follows...

We, like most companies are using Windows in an environment with a Domain Controller.  When the QNX GNU tools run, 
albiet make.exe, rm.exe, cd.exe, etc, there are thousands of packets that are sent to the domain controller.  I've used 
Wireshark to capture this behavior.  It seems that all the packets hitting the domain controller are some type of SMB 
authentication packet to the Domain Controller.

This really slows down the build on Windows especially if you have a slow connection, such as over a VPN.  Does anyone 
know what is going on here?

Thanks,
--john
Re: RE: Windows much slower then self-hosted  
> Is there any update for this problem?  I am also seeing this issue while 
> building on Windows, in this case it is Windows 7 Enterprise 64-bit.  I did 
> some digging into the problem and have narrowed the problem as follows...
> 
> We, like most companies are using Windows in an environment with a Domain 
> Controller.  When the QNX GNU tools run, albiet make.exe, rm.exe, cd.exe, etc,
>  there are thousands of packets that are sent to the domain controller.  I've 
> used Wireshark to capture this behavior.  It seems that all the packets 
> hitting the domain controller are some type of SMB authentication packet to 
> the Domain Controller.
> 
> This really slows down the build on Windows especially if you have a slow 
> connection, such as over a VPN.  Does anyone know what is going on here?
> 
> Thanks,
> --john

Funny that you`ve revived this thread.  I`m at a customer site, connected via VPN to the office and compilation has slow
 to a crawl.  I`ve done some digging and I too see the same thing that you are seeing.  If done some reasearch on the 
subject ( MSYS tools ) but came up empty.  Even at the office I can see slowdown some time, compiling big projets often 
show the CPUs not to be busy at 100% without a good explanation for it. I even tried moving Moments and the project to a
 RAM disk and it made no difference.  But if I disconnect from the network then thing pick up big time, but it`s still 
nowhere close to compilation on self hosted.




Re: RE: Windows much slower then self-hosted  
> Is there any update for this problem?  I am also seeing this issue while 
> building on Windows, in this case it is Windows 7 Enterprise 64-bit.  I did 
> some digging into the problem and have narrowed the problem as follows...
> 
> We, like most companies are using Windows in an environment with a Domain 
> Controller.  When the QNX GNU tools run, albiet make.exe, rm.exe, cd.exe, etc,
>  there are thousands of packets that are sent to the domain controller.  I've 
> used Wireshark to capture this behavior.  It seems that all the packets 
> hitting the domain controller are some type of SMB authentication packet to 
> the Domain Controller.
> 
> This really slows down the build on Windows especially if you have a slow 
> connection, such as over a VPN.  Does anyone know what is going on here?
> 
> Thanks,
> --john

I have finally found a solution to that problem.  In Windows/Preferences/Environment/C++/Build/Environment add the 
variable LOGONSERVER=\\\\YOURMACHINENAME and make sure you select the "Replace native environment with the specified one
".  You will even notice a decrease in compile time when you are connected to the company network.  Finally this enable 
usage of 100% cpu !!!!  

This is were I found the answer: 
http://readlist.com/lists/lists.sourceforge.net/mingw-users/1/9008.html
Re: RE: Windows much slower then self-hosted  
> > Is there any update for this problem?  I am also seeing this issue while 
> > building on Windows, in this case it is Windows 7 Enterprise 64-bit.  I did 
> 
> > some digging into the problem and have narrowed the problem as follows...
> > 
> > We, like most companies are using Windows in an environment with a Domain 
> > Controller.  When the QNX GNU tools run, albiet make.exe, rm.exe, cd.exe, 
> etc,
> >  there are thousands of packets that are sent to the domain controller.  
> I've 
> > used Wireshark to capture this behavior.  It seems that all the packets 
> > hitting the domain controller are some type of SMB authentication packet to 
> 
> > the Domain Controller.
> > 
> > This really slows down the build on Windows especially if you have a slow 
> > connection, such as over a VPN.  Does anyone know what is going on here?
> > 
> > Thanks,
> > --john
> 
> I have finally found a solution to that problem.  In Windows/Preferences/
> Environment/C++/Build/Environment add the variable LOGONSERVER=\\\\
> YOURMACHINENAME and make sure you select the "Replace native environment with 
> the specified one".  You will even notice a decrease in compile time when you 
> are connected to the company network.  Finally this enable usage of 100% cpu !
> !!!  
> 
> This is were I found the answer: 
> http://readlist.com/lists/lists.sourceforge.net/mingw-users/1/9008.html

I timed a make clean of our Makefile project.  It's made up of about 60 sub-projects which the root makefile recursively
 goes into.  With default setting clean time is 11 seconds, with LOGONSERVER changed ( or undefined ) 7 seconds, quite 
an improvement!!! 

I get shivers down my spine when I thing of all those under utilize CPUs ;-0

Merry Chrismast ;-)



.

Re: RE: Windows much slower then self-hosted  
> I timed a make clean of our Makefile project.  It's made up of about 60 sub-
> projects which the root makefile recursively goes into.  With default setting 
> clean time is 11 seconds, with LOGONSERVER changed ( or undefined ) 7 seconds,
>  quite an improvement!!! 
> 
> I get shivers down my spine when I thing of all those under utilize CPUs ;-0
> 

Build time went from 8:20 minutes to 6:18 minutes !!!!!!