I’ve used Pooch long ago to run a cluster of Macs but at that time the Machines I had were not very fast. I just downloaded and installed Pooch to cluster my Mac Mini and my Dual G5 PM. I ran a test of AltiVec Fractal. Using all three processors I got 6.8 Gigaflops! I’ll test it on a real world app later (like a video encoding)
Pooch

Cool!  Feel free to write a review about it in the Features section. :cool:

How does it compare to Xgrid?

Have you used Xgrid? I haven’t.

xphiler - 02 December 2005 04:12 PM

Using all three processors I got 6.8 Gigaflops!

Seems odd, you should have gotten something more like three times that result.
I got 6.5 GFLOPS with only one CPU, a 1.42 GHz G4.

Hmmm… let me reconfigure. My Mini is wired to my router and my G5 is wireless. We’ll see if it makes a difference if they are both wired.

Mikael - 02 December 2005 11:08 PM

xphiler - 02 December 2005 04:12 PM

Using all three processors I got 6.8 Gigaflops!

Seems odd, you should have gotten something more like three times that result.
I got 6.5 GFLOPS with only one CPU, a 1.42 GHz G4.

Interesting. I did some tweaking and managed 8.2 Gigaflops using the same network configuration. Pooch reports that I am using 3 nodes. My G5 is connected to my router using Airport Extreme, the Mini is connected using ethernet. Pooch willow allow me to use XGrid instead of SLP. I’ll test that next.

It seems as if only one processor is working, a G5.
This task should be ideal for distributed computing because of the simple SIMD instructions. Simple instructions to distribute to do a lot.

I’ve never used Xgrid but it’s in the firewall settings, and it seems to be really easy to configure as usual with Apple software.

yes xGrid looks easy but you need OS X server to run it. BTW 3 processors are detected and are being used. When I run the fractal with just the Mini I get 5 GFlops. When I run the fractal with Just the G5 (using both processors) I get 13 GFlops. You’d think I might get 18 GFlops together, no? Hmmmmm.....

I also changed the “Maximum Count” to something bigger, like 65536.
It’s as if it’s slow to get started.

We know that two CPU:s are not twice as fast, three are not three times as fast together. Maybe your processors should achieve 18 GFLOPS in theory but in practice only 16 GFLOPS which is not too bad.

Mikael - 10 December 2005 03:42 PM

I also changed the “Maximum Count” to something bigger, like 65536.
It’s as if it’s slow to get started.

We know that two CPU:s are not twice as fast, three are not three times as fast together. Maybe your processors should achieve 18 GFLOPS in theory but in practice only 16 GFLOPS which is not too bad.

I am reading the manual now. It seems that altivec fractal is not the best app to use to test on nodes with uneven performance. Fresnel explorer is better. Still reading so I’ll post again when I finish. You can use xGrid from Pooch if you want. I have not tested yet.

Here is Dean Dauger’s page if you’re interested. Dauger.com

I am able to get 15.2 GFlops by choosing Grid instead of MPI for processing. If you choose MPI Pooch waits for the last node to register and complete it’s job. That means the total time for the job is more than if you used one node alone.  Compare this to an alleged 35 GFlops on the latest Quad 2.5 Ghz G5.