HI
I'm new STELLAR -hpx.As i'm interested in participating GSOC 2017.And I'm
trying to learn what hpx is? and how it is implemented?.I watched Mr.Harmut
kaiser's parallelism in modern c++(video) in cppcon 2016.He explained about
partitioned vector as values that are not continues in memory and values
are chunk together at some parts and he also told that it distributes the
data over different computing nodes.And my doubt is if we chunk the data
together (for example : chunk d1 allocated for computing node n1,chunk d2
for node n2,d3 for n3 etc) how do we manage the situation if node n3 needs
the data d2 and d3 ? whether we should keep d2 allocated for both node n2
and node n3.If we are doing this we are actually duplicating data (d2) so
how  should we handle the situation.I'm not sure whether it is correct or
not one thing that I thought was we can have all the data that are needed
by only one node as a single partitioned vector.And if the data is needed
by multiple-nodes we can have that data as the separate vector and allow
the computing nodes to access the data through some sort of communication
between computing node and vector object.He also mentioned that to get a
better performance of GPU's we have to do direct manipulation of data on
GPU and if we use the communication between node and data it generally
decreases the performance. So how should we handle the situation?
Thanks you
Praveen velliengiri
_______________________________________________
hpx-users mailing list
[email protected]
https://mail.cct.lsu.edu/mailman/listinfo/hpx-users

Reply via email to