Parallel python in the cloud

[Charles Gagnon]

We were happily using PiCloud for several long calculations and we very happy with with it. With their realtime cores, we could take really large calculations set and run through fairly quickly.

Now that PiCloud is going away, we ran a few tests on Mutlyvac but so far, we are struggling to accomplish the same thing we had on PiCloud.

I have several "pieces" of my puzzle but can't seem to be able to put it together. I've seen and tried StarCluster and also various parallel python options but all options seem challenging to put together.

The goal is to mimic PiCloud, ie. loop through a function:

def some_NP_func(x, y):
...
return z

some_cloud.call(some_NP_func, a1, a2)

Which computes the function on the cloud. We use this often in for loops with arrays of arguments. The other scenario is:

some_cloud.map(some_NP_intense_func, [...], [...])

Which iterates through and returns results. We need to run a lot of this in batch from a scheduler so I always try to avoid interactive environment (how does iPython parallel work in batch?).

What is the preferred approach or method right now for heavy parallel computation like this?


Regards,

 

[Robert Kern]

We were happily using PiCloud for several long calculations and we very happy with with it. With their realtime cores, we could take really large calculations set and run through fairly quickly.

Now that PiCloud is going away, we ran a few tests on Mutlyvac but so far, we are struggling to accomplish the same thing we had on PiCloud.

I have several "pieces" of my puzzle but can't seem to be able to put it together. I've seen and tried StarCluster and also various parallel python options but all options seem challenging to put together.

The goal is to mimic PiCloud, ie. loop through a function:

def some_NP_func(x, y):
...
return z

some_cloud.call(some_NP_func, a1, a2)

Which computes the function on the cloud. We use this often in for loops with arrays of arguments. The other scenario is:

some_cloud.map(some_NP_intense_func, [...], [...])

Which iterates through and returns results. We need to run a lot of this in batch from a scheduler so I always try to avoid interactive environment (how does iPython parallel work in batch?).

IPython parallel works just fine "in batch". As far as your client code (i.e.
what you wrote above) is concerned, it's just another library. E.g.

https://github.com/ipython/ipython/blob/master/examples/Parallel Computing/nwmerge.py
https://github.com/ipython/ipython/blob/master/examples/Parallel Computing/itermapresult.py

etc.

--
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
that is made terrible by our own mad attempt to interpret it as though it had
an underlying truth."
-- Umberto Eco