[Paraview] How is builtin different from localhost?

2008-11-24 Thread Biao She
Hi all.
I have tested paraview on a builtin server and a localhost server(run
pvserver and connect client to it on the same computer). It seems to me that
the builtin server is much faster than localhost server. Since the computer
is the same, i am wondering if the localhost server have to readback all the
data in GPU and transfer these data to client in order to be displayed? On
the other hand, for the builtin server, the data in GPU is displayed
directly(no readback). Could you please explain why the localhost is slower
than builtin?
Thanks very much!

Aaron
___
ParaView mailing list
ParaView@paraview.org
http://www.paraview.org/mailman/listinfo/paraview


Re: [Paraview] How is builtin different from localhost?

2008-11-24 Thread David E DeMarle
On Mon, Nov 24, 2008 at 1:33 PM, Biao She [EMAIL PROTECTED] wrote:
 Hi all.
 I have tested paraview on a builtin server and a localhost server(run
 pvserver and connect client to it on the same computer). It seems to me that
 the builtin server is much faster than localhost server. Since the computer
 is the same, i am wondering if the localhost server have to readback all the
 data in GPU and transfer these data to client in order to be displayed? On
 the other hand, for the builtin server, the data in GPU is displayed
 directly(no readback). Could you please explain why the localhost is slower
 than builtin?
 Thanks very much!

 Aaron

 ___
 ParaView mailing list
 ParaView@paraview.org
 http://www.paraview.org/mailman/listinfo/paraview



The builtin server lives inside the same process as the client. All
communication between server and client takes place via pointers and
vtk class methods are directly called.

Localhost ie, running pvserver on the same machine as the client and
connecting to it consists of two separate processes. It is slower
because both processes compete for CPU cycles and memory space more
importantly, because communication between the two takes place via TCP
sockets.

-- 
David E DeMarle
Kitware, Inc.
RD Engineer
28 Corporate Drive
Clifton Park, NY 12065-8662
Phone: 518-371-3971 x109
___
ParaView mailing list
ParaView@paraview.org
http://www.paraview.org/mailman/listinfo/paraview


Re: [Paraview] How is builtin different from localhost?

2008-11-24 Thread Biao She
Thanks for your reply.
I have another question though. You said the two processes communicated via
TCP sockets. What do they actually transfer? The final rendered image? The
vtk class methods calls? Or both?
I also have tested the pvserver on a remote computer, and I connected to the
server with 100 M network. I am wondering if the network speed is good
enough for paraveiw? In other words, no huge display delay because of
network?
Thanks.

Biao

On Mon, Nov 24, 2008 at 11:43 AM, David E DeMarle
[EMAIL PROTECTED]wrote:

 On Mon, Nov 24, 2008 at 1:33 PM, Biao She [EMAIL PROTECTED] wrote:
  Hi all.
  I have tested paraview on a builtin server and a localhost server(run
  pvserver and connect client to it on the same computer). It seems to me
 that
  the builtin server is much faster than localhost server. Since the
 computer
  is the same, i am wondering if the localhost server have to readback all
 the
  data in GPU and transfer these data to client in order to be displayed?
 On
  the other hand, for the builtin server, the data in GPU is displayed
  directly(no readback). Could you please explain why the localhost is
 slower
  than builtin?
  Thanks very much!
 
  Aaron
 
  ___
  ParaView mailing list
  ParaView@paraview.org
  http://www.paraview.org/mailman/listinfo/paraview
 
 

 The builtin server lives inside the same process as the client. All
 communication between server and client takes place via pointers and
 vtk class methods are directly called.

 Localhost ie, running pvserver on the same machine as the client and
 connecting to it consists of two separate processes. It is slower
 because both processes compete for CPU cycles and memory space more
 importantly, because communication between the two takes place via TCP
 sockets.

 --
 David E DeMarle
 Kitware, Inc.
 RD Engineer
 28 Corporate Drive
 Clifton Park, NY 12065-8662
 Phone: 518-371-3971 x109




-- 
She, Biao
Department of Computing Science,
University of Alberta, Edmonton, Canada
___
ParaView mailing list
ParaView@paraview.org
http://www.paraview.org/mailman/listinfo/paraview


Re: [Paraview] How is builtin different from localhost?

2008-11-24 Thread David E DeMarle
On Mon, Nov 24, 2008 at 2:22 PM, Biao She [EMAIL PROTECTED] wrote:
 Thanks for your reply.
 I have another question though. You said the two processes communicated via
 TCP sockets. What do they actually transfer? The final rendered image? The
 vtk class methods calls? Or both?
 I also have tested the pvserver on a remote computer, and I connected to the
 server with 100 M network. I am wondering if the network speed is good
 enough for paraveiw? In other words, no huge display delay because of
 network?
 Thanks.

 Biao

 On Mon, Nov 24, 2008 at 11:43 AM, David E DeMarle [EMAIL PROTECTED]
 wrote:

 On Mon, Nov 24, 2008 at 1:33 PM, Biao She [EMAIL PROTECTED] wrote:
  Hi all.
  I have tested paraview on a builtin server and a localhost server(run
  pvserver and connect client to it on the same computer). It seems to me
  that
  the builtin server is much faster than localhost server. Since the
  computer
  is the same, i am wondering if the localhost server have to readback all
  the
  data in GPU and transfer these data to client in order to be displayed?
  On
  the other hand, for the builtin server, the data in GPU is displayed
  directly(no readback). Could you please explain why the localhost is
  slower
  than builtin?
  Thanks very much!
 
  Aaron
 
  ___
  ParaView mailing list
  ParaView@paraview.org
  http://www.paraview.org/mailman/listinfo/paraview
 
 

 The builtin server lives inside the same process as the client. All
 communication between server and client takes place via pointers and
 vtk class methods are directly called.

 Localhost ie, running pvserver on the same machine as the client and
 connecting to it consists of two separate processes. It is slower
 because both processes compete for CPU cycles and memory space more
 importantly, because communication between the two takes place via TCP
 sockets.

 --
 David E DeMarle
 Kitware, Inc.
 RD Engineer
 28 Corporate Drive
 Clifton Park, NY 12065-8662
 Phone: 518-371-3971 x109



 --
 She, Biao
 Department of Computing Science,
 University of Alberta, Edmonton, Canada


It depends on the settings.

* vtk class method names and arguments to data processing filters
(clip, slice, contour, surface generation) are always sent from the
client to the server whenever a filter parameter changes.

* If the resulting geometry is smaller than the
settings-remote-server-remote render threshold, than the geometry
is sent back (on each filter parameter change, NOT on every camera
setting change) to the client and rendered locally.

* If the geometry is larger than that, then the image is rendered by
the server and the pixels are sent back every frame.

* If the visible geometry is small enough that the client can render
it interactively, than a 100M connection shouldn't be any problem.

-- 
David E DeMarle
Kitware, Inc.
RD Engineer
28 Corporate Drive
Clifton Park, NY 12065-8662
Phone: 518-371-3971 x109
___
ParaView mailing list
ParaView@paraview.org
http://www.paraview.org/mailman/listinfo/paraview