[julia-users] starting a worker limits blas to one thread

[Michael Eastwood]

Hello julia-users,

I've noticed that if I start a worker blas is all of the sudden restricted
to only running on one thread. See for example the following log. Here I
multiply a large matrix by itself, add a remote worker, and repeat the
multiplication (on the master process). In the latter case blas only uses
one thread despite the fact it was happily using 16 threads before adding
the worker. (Note that this log only shows timings but I've verified the
number of threads being used by watching top).

Can this behavior be circumvented?

$ julia
               _
   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "?help" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.4.3 (2016-01-12 21:37 UTC)
 _/ |\__'_|_|_|\__'_|  |
|__/                   |  x86_64-unknown-linux-gnu

julia> function multiply(A)
           A'*A
       end
multiply (generic function with 1 method)

julia> A = randn(5000, 5000);

julia> ENV["OMP_NUM_THREADS"]
"16"

julia> @time multiply(A); # 16 threads used as expected
  1.302345 seconds (634.91 k allocations: 220.681 MB, 3.01% gc time)

julia> addprocs([("astm10", 1)]) # add a process on a remote machine
1-element Array{Int64,1}:
 2

julia> ENV["OMP_NUM_THREADS"] # unchanged
"16"

julia> @time multiply(A); # only one thread used
 10.126372 seconds (9 allocations: 190.735 MB, 0.02% gc time)

julia> versioninfo()
Julia Version 0.4.3
Commit a2f713d* (2016-01-12 21:37 UTC)
Platform Info:
  System: Linux (x86_64-unknown-linux-gnu)
  CPU: Intel(R) Xeon(R) CPU E5-2630 v3 @ 2.40GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT NO_AFFINITY NEHALEM)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.3


Thanks,
Michael