Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > comp.lang.clipper.visual-objects > #2786 > unrolled thread

visual objects network performance problem

Started by"Paul Harrison" <stable_eyes@msn.com.au>
First post2012-01-08 19:37 +1100
Last post2012-01-19 00:13 +0100
Articles 7 — 7 participants

Back to article view | Back to comp.lang.clipper.visual-objects


Contents

  visual objects network performance problem "Paul Harrison" <stable_eyes@msn.com.au> - 2012-01-08 19:37 +1100
    Re: visual objects network performance problem "Willie Moore" <williem@wmconsulting.com> - 2012-01-08 14:13 -0600
    Re: visual objects network performance problem Mathias <mathias.hakansson@consultec.se> - 2012-01-08 22:58 -0800
    Re: visual objects network performance problem "Dirk (Belgium)" <dirk.dot.herijgers@pbprojects.dot.be> - 2012-01-09 08:42 +0000
    Re: visual objects network performance problem Sherlock <sherlock@sherlock.com.au> - 2012-01-13 07:15 -0800
    Re: visual objects network performance problem Martin <spam@spam.spam> - 2012-01-18 11:09 +0000
      Re: visual objects network performance problem "Marc Verkade [Marti]" <marcatm@rti.nl> - 2012-01-19 00:13 +0100

#2786 — visual objects network performance problem

From"Paul Harrison" <stable_eyes@msn.com.au>
Date2012-01-08 19:37 +1100
Subjectvisual objects network performance problem
Message-ID<cQcOq.1304$v14.348@viwinnwfe02.internal.bigpond.com>
I have a network performance problem with a visual objects application. I am 
hoping there is somebody who can explain what is happening so that I can 
solve the problem.

On a terminal where the application and the data are pulled from the server 
it takes 3-4 secs to read and report on 8500 records of a dbf server (cdx). 
To do the same on the server takes about the same time. So there is little 
overhead imposed by the network.

If the user on the server has the same server open then the same operation 
on the terminal as above now takes 27 secs. If the user on the server closes 
the server (window) the above operation on the terminal still takes 27 secs. 
I should at this point tell you the server is opened in shared mode, 
readwrite with concurrencycontrol set to ccoptimistic. The application is 
only processing the records to generate a report (no writing to the file). 
It also makes no difference if we turn off notification on the server.

One further interesting observation is that before processing the report we 
close the server and then reopen it (using self:use(dbserver) on the window 
(clears client list of the server ?). This appears to clear the effect of 
the server having been open on the server, that is the report takes 3-4 secs 
again.

I cannot understand why moving through server records takes upto 10 times 
longer on a terminal if the server is also open on the server.

Any information would be welcome and appreciated.


Paul Harrison 

[toc] | [next] | [standalone]


#2787

From"Willie Moore" <williem@wmconsulting.com>
Date2012-01-08 14:13 -0600
Message-ID<jectc7$rbv$1@speranza.aioe.org>
In reply to#2786
Paul,

what version of VO?

Regards,
Willie
Vulcan VIP

[toc] | [prev] | [next] | [standalone]


#2788

FromMathias <mathias.hakansson@consultec.se>
Date2012-01-08 22:58 -0800
Message-ID<33284690.25.1326092285979.JavaMail.geo-discussion-forums@yqnt2>
In reply to#2786
There is an (old) explanation on this phenomena here http://www.superbase.com/services_tech_support_oplocks.htm.

The performance is better now with windows server 2008 together with Windows 7, as this combination of OSs can change the packet size dynamically, that is if there isn't a XP machine on the network as well. So reading the old texts from 2002 gives us an conslution that is still true;

* Microsoft wants us to stop using file sharing databases.
* If we do we will be punished with bad performance.

There are a couple of things you can do to make your application perform better.
* If you want to show a database in a browser
    - Don't use filters, use indexes with conditions instead.
    - Try to keep the record size down (use sub tables for complementary data).
    - Try to keep the file size down (you could move old records to another database).

/Mathias

[toc] | [prev] | [next] | [standalone]


#2789

From"Dirk (Belgium)" <dirk.dot.herijgers@pbprojects.dot.be>
Date2012-01-09 08:42 +0000
Message-ID<jee99b$gu8$1@dont-email.me>
In reply to#2786
Paul,

The time of 27 seconds is the 'real' time for accessing the data over
the network, when no caching is performed.

However, when 'windows' see that there is only 1 computer accessing the
dbf-file, it pushes the whole file to this computer, so access takes
about 4 seconds on your 'terminal'.

As soon as another user on another 'terminal' or server also opens the
same data, the cache is deleted on the terminal-computer, and real
access over the network has to be done.

The principle of caching exist already very long (win95 - win2000 -
winXP - etc) but from Windows VISTA (and win7) the caching is very
efficient. This is done by 'SMB2'. WinXP and before are using 'SMB'
which is less efficient.

Dirk

Paul Harrison wrote:

> I have a network performance problem with a visual objects
> application. I am hoping there is somebody who can explain what is
> happening so that I can solve the problem.
> 
> On a terminal where the application and the data are pulled from the
> server it takes 3-4 secs to read and report on 8500 records of a dbf
> server (cdx). To do the same on the server takes about the same time.
> So there is little overhead imposed by the network.
> 
> If the user on the server has the same server open then the same
> operation on the terminal as above now takes 27 secs. If the user on
> the server closes the server (window) the above operation on the
> terminal still takes 27 secs. I should at this point tell you the
> server is opened in shared mode, readwrite with concurrencycontrol
> set to ccoptimistic. The application is only processing the records
> to generate a report (no writing to the file). It also makes no
> difference if we turn off notification on the server.
> 
> One further interesting observation is that before processing the
> report we close the server and then reopen it (using
> self:use(dbserver) on the window (clears client list of the server
> ?). This appears to clear the effect of the server having been open
> on the server, that is the report takes 3-4 secs again.
> 
> I cannot understand why moving through server records takes upto 10
> times longer on a terminal if the server is also open on the server.
> 
> Any information would be welcome and appreciated.
> 
> 
> Paul Harrison



-- 

[toc] | [prev] | [next] | [standalone]


#2805

FromSherlock <sherlock@sherlock.com.au>
Date2012-01-13 07:15 -0800
Message-ID<24d027cb-fac2-4a91-bc71-89ce73088358@c1g2000pbf.googlegroups.com>
In reply to#2786
Paul

You could try this...

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\services\LanmanServer
\Parameters]
"smb2"=dword:00000000

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\services\LanmanWorkstation
\Parameters]
"FileInfoCacheLifetime"=dword:00000000
"FileNotFoundCacheLifetime"=dword:00000000
"DirectoryCacheLifetime"=dword:00000000

Phil McGuinness
-------

[toc] | [prev] | [next] | [standalone]


#2843

FromMartin <spam@spam.spam>
Date2012-01-18 11:09 +0000
Message-ID<hLxRq.2$dE1.0@newsfe15.ams2>
In reply to#2786
Is it worth using ADS?

On 08/01/2012 08:37, Paul Harrison wrote:
> I have a network performance problem with a visual objects application. I am
> hoping there is somebody who can explain what is happening so that I can
> solve the problem.
>
> On a terminal where the application and the data are pulled from the server
> it takes 3-4 secs to read and report on 8500 records of a dbf server (cdx).
> To do the same on the server takes about the same time. So there is little
> overhead imposed by the network.
>
> If the user on the server has the same server open then the same operation
> on the terminal as above now takes 27 secs. If the user on the server closes
> the server (window) the above operation on the terminal still takes 27 secs.
> I should at this point tell you the server is opened in shared mode,
> readwrite with concurrencycontrol set to ccoptimistic. The application is
> only processing the records to generate a report (no writing to the file).
> It also makes no difference if we turn off notification on the server.
>
> One further interesting observation is that before processing the report we
> close the server and then reopen it (using self:use(dbserver) on the window
> (clears client list of the server ?). This appears to clear the effect of
> the server having been open on the server, that is the report takes 3-4 secs
> again.
>
> I cannot understand why moving through server records takes upto 10 times
> longer on a terminal if the server is also open on the server.
>
> Any information would be welcome and appreciated.
>
>
> Paul Harrison
>
>

[toc] | [prev] | [next] | [standalone]


#2850

From"Marc Verkade [Marti]" <marcatm@rti.nl>
Date2012-01-19 00:13 +0100
Message-ID<4f175241$0$796$58c7af7e@news.kabelfoon.nl>
In reply to#2843
With more than 2 simultanious users?
Yes!
Greetz, Marc

"Martin"  schreef in bericht news:hLxRq.2$dE1.0@newsfe15.ams2...

Is it worth using ADS?

On 08/01/2012 08:37, Paul Harrison wrote:
> I have a network performance problem with a visual objects application. I 
> am
> hoping there is somebody who can explain what is happening so that I can
> solve the problem.
>
> On a terminal where the application and the data are pulled from the 
> server
> it takes 3-4 secs to read and report on 8500 records of a dbf server 
> (cdx).
> To do the same on the server takes about the same time. So there is little
> overhead imposed by the network.
>
> If the user on the server has the same server open then the same operation
> on the terminal as above now takes 27 secs. If the user on the server 
> closes
> the server (window) the above operation on the terminal still takes 27 
> secs.
> I should at this point tell you the server is opened in shared mode,
> readwrite with concurrencycontrol set to ccoptimistic. The application is
> only processing the records to generate a report (no writing to the file).
> It also makes no difference if we turn off notification on the server.
>
> One further interesting observation is that before processing the report 
> we
> close the server and then reopen it (using self:use(dbserver) on the 
> window
> (clears client list of the server ?). This appears to clear the effect of
> the server having been open on the server, that is the report takes 3-4 
> secs
> again.
>
> I cannot understand why moving through server records takes upto 10 times
> longer on a terminal if the server is also open on the server.
>
> Any information would be welcome and appreciated.
>
>
> Paul Harrison
>
> 

[toc] | [prev] | [standalone]


Back to top | Article view | comp.lang.clipper.visual-objects


csiph-web