Parallel Computing Planning Issue

[alanedelman]

As we talk about parallel computing in various contexts, it might be handy to keep in mind all the high level moving parts. Here are some that come to mind, just to get the ball rolling.

Hardware

• [ ] Shared Memory Machines (Threads: See #1790, #1802)
• [ ] Distributed Memory (especially networks of shared memory) (spawns, MPI?, migrating threads?)
• [ ] GPUs? (we do/don't believe these are here to stay)

Tools

• [ ] Nice Graphical Performance Tools and Instrumentation (My first wish: just have Green (working) / Red (idle) on every processor easily every time I do parallel julia)

Programming Models

• [ ] Persistent Data as in Global Array Syntax Darrays (Star-P and others)
• [ ] CILK/Spawns
• [ ] Map/reduce etc.

Communication Layers

• [ ] ZMQ, MPI, sockets

Schedulers

• [ ] ??
• [ ] Does one master know all state?
• [ ] Can our serial model be in conflict with various parallel models?

Libraries

• [ ] Scalapack, parallel FFTW?, sparse matrices, PETSc , (star-p like to compare with more pure julia methods?)

[ViralBShah]

I am closing this as very generic. A bunch of specific things are being done already.

[classner]

Where will the status of the threading branch be tracked? And what is its current status? That one was very promising! :)

[ViralBShah]

1790