I guess without more knowledge of what your problem space really is I can’t
understand why the problem doesn’t decompose into small problems which can be
farmed out to a java space. Certainly, there are issues if you are using a
C-language application on and MDI platform for parallelism. If
Regarding multicast RMI:
When a node joins the multicast group and begins receiving remote method
invocations, it will need to be able to start accepting method
invocations from the correct point in the stream, that is from the top
level object in the remote method invocation.
When I was
Gregg,
Graphs traversal is in general a non-local problem with irregular, data
dependent parallelism (the available fine grained parallelism for a vertex
depends on its edge list, the size of the edge list can vary over many
orders of magnitude, edges may connect vertices that are non local, and
Why doesn’t java spaces let you submit requests and have them worked on and
results returned without assigning any particular node any specific
responsibility?
Gregg
> On Aug 1, 2015, at 12:06 PM, Bryan Thompson wrote:
>
> First, thanks for the responses and the interest in
I’ve wondered about doing this in the past, but for the workloads I’ve worked
with, I/O time has been relatively low compared to processing time. I’d guess
there’s some combination of message frequency, cluster size and message size
that makes it compelling.
The idea is interesting, though,
Hello,
I am wondering if anyone has looked into creating tree based algorithms for
multi-cast of RMI messages for river. Assuming a local cluster, such
patterns generally have log(p) cost for a cluster with p nodes.
For the curious, this is how many MPI messages are communicated under the
hood.