I'm using a large number of shared arrays and storing their references in a
dictionary. I've noticed when I end up having to do a remotecall on all the
workers that just passing around the references is pretty slow. In the
example below it generally takes between 1/2 to a full second to finish all
the remotecalls on 7 workers. I'm using Julia version 0.4.1 and CentOS 7.
addprocs(7)
@everywhere function
test_shared(i::Dict{Tuple{Int64,Int64},SharedArray{Float64}})
println("test_shared $(myid()) time: $(Dates.unix2datetime(time()))")
end
function main()
b = Dict{Tuple{Int64,Int64},SharedArray{Float64}}()
for x = 1:3
for y=1:300
b[x,y]=SharedArray(Float64, 1, init=zeros(Float64, 1))
end
end
for x = 1:nprocs()
if x != myid()
@time remotecall(x,test_shared,b)
end
end
end
main()
sleep(2)
main()
sleep(2)
