Valentin Churavy
Valentin Churavy
You have to install the same Julia version on the Cluster. You can change which version of Julia is being started, but it has to be the same one accross...
I don't know hoe the R package does that, but as long as you have `qsub` installed we should be able to do something similar. The only caveat is that...
Yes I think the SPMD code might actually be a good example for a separate package, (but still in the same repository.)
Hm, I think Distributed assumes that we can can have a connection for each direction. I don't think there is a hard reason for the implementation detail, except that it...
I consider this issue rather speculative, but we do need somewhere to issue operations without them necessarily blocking on each other. CuArray stream interface is particularly interesting, but relies on...
Jup that is a bug in the constructor: https://github.com/JuliaParallel/DistributedArrays.jl/blob/e6915d28b59f564043597985bd5af1896f3e380b/src/darray.jl#L110 `localpartindex(pids)` assumes that there is only one chunk on a process and will return the first one.
There is https://support.hdfgroup.org/HDF5/PHDF5/, but that is intended for a MPI environment. Parallel/distributed IO depends a lot on what filesystem you are using. As an example Lustre offers enough consistency guarantees...
It is now: https://github.com/JuliaParallel/DagScheduler.jl
Yes we are and this is how we work around it... https://github.com/JuliaGPU/GPUArrays.jl/blob/a305f78d8cd7310c0fa07310e8974d5350cc293b/src/abstractarray.jl#L53-L74
cc: @tkf