Amit Murthy

What we are looking for here I think is - slurm to only allocate compute nodes (and _not_ launch any julia workers) - write node names to a hostfile -...

This is because `remote_do` does not return any response to the caller. In your example above: `remotecall` will result in an error when the response is fetched unlike in the...

That block of code is an optimization to avoid serializing global objects if unchanged. Relevant especially in the case of largish arrays. Short of removing the optimization and serializing all...

No, `object_id` of `x` is unchanged in your example above - it is the same object. The global is serialized if either `object_id` or the hash value of the binding...

The following should do what you need (no batching and retrying though): ``` v = initial_val for x in Base.AsyncGenerator(y->remotecall_fetch(foo_map_func, default_worker_pool(), y), collection_to_be_mapreduced; ntasks=nworkers()) v = foo_reduce_func(v, x) end ```...

This would be a tricky situation to address if the RNG is being serialized/deserialized every time. On 0.6, `global const rng = MersenneTwister(3742)` would ensure consistency with/without additional workers.

mutables are not synchronized across processes in any manner. There is no cluster wide shared state for mutables referenced in closures. A common data store external to the julia processes...

For performance reasons remote calls executing locally short-circuit the entire serialization-deserialization cycle of the closure. I get your view about consistent results between nprocs=1 and nprocs>1 in absolutely all cases....

I wonder how and why JuliaLang/julia#22588 affected worker startup time. @vtjnash ?

@andreasnoack / @ViralBShah care to comment on the interface for lazy connection setup in JuliaLang/julia#22814?