[Starpu-devel] About the prediction model of data transfer time

[Lennox Wu <lennox.wu@gmail.com>]

Hi all,
I am interested in that how does StarPU predict the transfer time.
I didn't find the prediction model from your papers(Was it written 
somewhere?),
so I tried to find it from the source code.
I  found the function  "starpu_transfer_predict", is it the prediction model?
I can't understand the meaning behind the fomular
"2*(topology->ncudagpus+topology->nopenclgpus)",
could anybody explain it?

double starpu_transfer_predict(unsigned src_node, unsigned dst_node,
size_t size)
{
        double bandwidth = bandwidth_matrix[src_node][dst_node];
        double latency = latency_matrix[src_node][dst_node];
        struct _starpu_machine_topology *topology =
&_starpu_get_machine_config()->topology;

        return latency +
(size/bandwidth)*2*(topology->ncudagpus+topology->nopenclgpus);
}


Best,
Lennox