SSH as an out of the box option?

[cozymandias]

First of all, I'm really glad to see that there is already some R work going on with this. I ordered a couple boards yesterday with the hope that it would be possible to get them working with R. Then I came across this set of posts and was really glad to see that a lot of work has already been done.

Second, wouldn't it be possible to get things working "out of the box" using SSH? Maybe not the highest performance solution, but something that would work with one of the doXXXX pacakges (i.e., doParallel) would make this really easy for general purpose use.

I'm thinking about this based on how I build a cluster of laptops at my house when I need a few extra cores, maybe I'm missing something about the nature of the Parallella structure and/or cores.

[timpart]

[censix]

cozymandias, you have to keep in mind that, as things are currently, the only way to really use the E16/64 from R is via the clunky OpenCL example I posted here, or maybe by using inline C code in R, which I have not tried. There is no direct interface library yet. So at the moment, if you want to build a small cluster with several cards and run R on each instance, you can of course, but it would just sum up to using N (=number of cluster nodes) Zynq7020 processors and no E16/64 acceleration.

[cozymandias]

@censix, thanks for the clarification. Do you know what is currently the most promising route to the direct interface library? Given that this must be a general desire, I imagine someone is already working on it.

[cozymandias]

Just saw the MPI lite thread. Looks like there is some momentum there.