intermittent "wrong number of arguments" in parallel

[alanedelman]

I've been loving running a parallel histogram on up to 75 processors lately.
I'd say once out of every four or five runs, I get some havoc that seems to be
related to the handover of messages from remote processors.

I will show

1. the code working

2. the code breaking

3. the code

4. the code working

julia> include("demodriver.jl")
Matrix Size:25
Number of Monte Carlo trials (in millions): 0.164
Trials/Processor: 4000
Number of Processors: 41

2)the code breaking

    From worker 33:  in anonymous at multi.jl:416
    From worker 34: exception on 34: ERROR: in anonymous: wrong number of arguments
    From worker 34:  in anonymous at multi.jl:1182
    From worker 34:  in anonymous at multi.jl:1142
    From worker 34:  in anonymous at multi.jl:416
    From worker 35: exception on 35: ERROR: in anonymous: wrong number of arguments
    From worker 35:  in anonymous at multi.jl:1182
    From worker 35:  in anonymous at multi.jl:1142
    From worker 35:  in anonymous at multi.jl:416
    From worker 37: exception on 37: ERROR: in anonymous: wrong number of arguments
    From worker 37:  in anonymous at multi.jl:1182
    From worker 37:  in anonymous at multi.jl:1142
    From worker 37:  in anonymous at multi.jl:416
    From worker 36: exception on 36: ERROR: in anonymous: wrong number of arguments
    From worker 36:  in anonymous at multi.jl:1182
    From worker 36:  in anonymous at multi.jl:1142
    From worker 36:  in anonymous at multi.jl:416
    From worker 39: exception on 39: ERROR: in anonymous: wrong number of arguments
    From worker 39:  in anonymous at multi.jl:1182
    From worker 39:  in anonymous at multi.jl:1142
    From worker 39:  in anonymous at multi.jl:416
    From worker 40: exception on 40: ERROR: in anonymous: wrong number of arguments
    From worker 40:  in anonymous at multi.jl:1182
    From worker 40:  in anonymous at multi.jl:1142
    From worker 40:  in anonymous at multi.jl:416
    From worker 41: exception on 41: ERROR: in anonymous: wrong number of arguments
    From worker 41:  in anonymous at multi.jl:1182
    From worker 41:  in anonymous at multi.jl:1142
    From worker 41:  in anonymous at multi.jl:416
    From worker 38: exception on 38: ERROR: in anonymous: wrong number of arguments
    From worker 38:  in anonymous at multi.jl:1182
    From worker 38:  in anonymous at multi.jl:1142
    From worker 38:  in anonymous at multi.jl:416
ERROR: no method +(Array{Int64,1},ErrorException)
 in mapreduce at reduce.jl:118
 in preduce at multi.jl:1245
 in include_from_node1 at loading.jl:82
at /home/edelman/julia/borodingorin/demodriver.jl:1329

3. the code

==== demodriver.jl ======

@everywhere include("demo.jl");

n=25
t=4000
x=[0:.005:0.75]

p=nprocs()
println("Matrix Size:",n)
println("Number of Monte Carlo trials (in millions): ", p*t/1e6)
println("Trials/Processor: ",t)
println("Number of Processors: ",p)

z=@parallel (+) for i=1:p 
    demo(n,t,x)
end 

x=x[2:end]-(x[2]-x[1])/2
y=z/(p*t)/(x[2]-x[1])

==== demo.jl================

function demo(n,t,x)
v=zeros(t);
for i=1:t
   a=randn(n,n);
   v[i]=(eigvals(a'*a))[3];
end
(x,y)=hist(v,x)
return(y)
end

[pao: formatting]

[staticfloat]

Can I assume this is run on julia.mit.edu? How recent is your Julia build?

[ViralBShah]

Yes, it is on julia.mit.edu.

@alanedelman Can you run this with the latest julia?

[JeffBezanson]

I can still get this to happen intermittently.