Re: MathLink Speed.
- To: mathgroup at smc.vnet.net
- Subject: [mg22611] Re: MathLink Speed.
- From: Jens-Peer Kuska <kuska at informatik.uni-leipzig.de>
- Date: Tue, 14 Mar 2000 22:46:25 -0500 (EST)
- Organization: Universitaet Leipzig
- References: <8aekm0$3eu@smc.vnet.net>
- Sender: owner-wri-mathgroup at wolfram.com
Hi,
you are doing nothing wrong. But I don't know what
you measure. Timing[] measure only the time of the
kernel to create the data and to prepare the data.
You can't just compute 1800 * 4 * sizeof(double) to get the
memory. You have atleast the "Polygon" heads take into account
and the and the "List" heads for the point list, the "List" heads
of the coordinates as well as the integers for count the number of
points and coordinates.
The only high level protocol that is faster than MathLink is MPI
but it is not able to transfer *any* Mathematica structure.
Regards
Jens
The speed of MathLink is less i
Francois Guimbretiere wrote:
>
> Hi All,
> I am using MathLink to transfer Graphics3D object to a high resolution
> experimental display and I am somewhat surprised by the speed of the link. For
> example, it takes 1.1 seconds to transfer the graphic object generated by
> SphericalPlot3D[1,{theta,0,Pi,Pi/20},{phi,0,Pi,Pi/35}]. This graphic object
> contains about 1800 polygons. (For this test, I was running both Mathematica 4.0
> for Student and the external program on a quad-processor machine NT 4.0. The
> link was created using -linkcreate on the external program side). My
> understanding is that by default the link is using FileMap. Doing the math, it
> seems that the bandwidth is extremely low for this kind of RPC. Are these
> results typical? Am I doing something wrong?
> Is there any trick that can be used to speed up MathLink protocol?
>
> Francois.