@Nasser It seems there is no really one way to do things in Python ;) ... According to [this "hot network question"] (stackoverflow.com/questions/30451252/…) yesterday, it seems Python users have the same kind of questions than here on MSE ... I can't remember the last time I made a for loop in mma ... I was tempted to post an answer Wolfram Language which in that case seems to me far more readable ;)

Actually, I'd like (to have time) to learn a new language (besides mma and C) and I wonder if people views have changed since these posts (mathematica.stackexchange.com/questions/25702/…). The use of a (mma like) notebook in Python and Julia seem very interesting. On the other side, Javascript is everywhere ... and Java seem so natural to use with mma ...

@TheToad another one: mathematica.stackexchange.com/questions/48222/…

@Szabolcs I totally agree. It should be enabled and sandboxed by default, not disabled. Disabling it is a quick and dirty hack that I think broadly speaking is not useful for the user or for security.

@Sektor are you? I felt it is of limited usefulness, though certainly interesting as a toy.

Harder to make than it seems! (Generate random circles; start from a random point on every circle expanding the arc at constant speed; stop when intersection with another arc is met.)

I'm using the parallel tools with RunProcess. Subkernels regularly crash, but the problem is random and hard to reproduce. Has anyone else seen this?

@Szabolcs Yes, chat.stackexchange.com/transcript/2234?m=21763632#21763632

@OleksandrR. I have 5 of those; bought them last summer, but only last week I had the time to sit down and boot it up, fool around with it. I started to read the manual and hopefully in the coming week will finish building this mini-cluster

@OleksandrR. Why are you not impressed ?

Another RunProcess bug: it doesn't close open file handles. On OS X, if you try the following (DON'T!), Do[RunProcess["date","StandardOutput"],{20000}], Mathematica starts issuing "Too many open files in system" messages and generally everything in the OS breaks ... so don't try it if you have unsaved work.

My grave accent key died yesterday for no apparent reason. So if my comments/posts don't look as good as they should, you know why... :'(

@OleksandrR. @TheToad I opened an official question:

@SjoerdC.deVries quite...

If anyone has tried MATLink on OSX with MATLAB_R2014b on OSX? I manually set the path with MATLAB=/Applications/MATLAB_R2014b.app and I can run mengine.sh in the terminal, but still the OpenMATLAB[] hangs.

@OleksandrR. Plus, we'll see if it turns out to be something feasible to work with supercomputer.io/#

@Pickett You should ping @Szabolcs and @TheToad about this.

@Szabolcs and @TheToad the debug information requested on the website is "MATLink 1.1... 10.1.0 for Mac OS X x86 (64-bit)..." /bin/csh is correct. I can't recompile using the built-in command, haven't tried manually since I don't know if it will help or not (only recommended for Linux).

@Pickett Can you open MATLAB by itself (not via MATLink)? Does it need a license file/VPN to run and if so, did you try using MATLink via the same network? Do you have multiple versions of MATLAB?

OS X should run fine out of the box and shouldn't need compiling. You need to change mengine.sh only if you want to use a specific version of MATLAB or if it's not in /Applications/.... R2014b is a good one (R2014a had some non-fixable issues)

Could you please try trashing the entire MATLink dir and start with a fresh download from the website?

@TheToad I downloaded a new set of files from the website and tried again, but it's the same thing. The reason I changed the version explicitly before is because I have several versions installed, including R2014a, and wasn't sure which one it was picking. I'm running MATLAB locally on my computer, it has been activated with a valid license. Never asks me to verify the license or anything like that.

@TheToad Should MATLAB be running or not when one does OpenMATLAB[]?

@Pickett OpenMATLAB[] will launch a new headless MATLAB process. If your license allows you to run only 1 at a time, this could be a problem if you already have one open.

@TheToad ok, I've tried both. I also think it allows for more than that (student license).

And when you said you ran mengine.sh in the terminal, it successfully showed you Create link: at the end?

@TheToad yes

Hmmm.

Ok, I don't have Mathematica or MATLAB (or MATLink) on this machine and I'm mostly going by memory, but could you try creating a link at the prompt, say foo and then load MATLink with Needs and then instead of OpenMATLAB, do ConnectEngine["foo"]?

All of these in a fresh session please

Oh, and could you also past the output of MATLink`Developer`ShowLog[]?

matlink Closed 0 old link objects.

@TheToad I'm not sure what "link at the prompt" means.

or maybe I get what it is but I don't know how to create a link to MATLAB.

Some more error messages from earlier today here.

@Pickett I mean when you run mengine.sh in a terminal and it says "Create link:" and waits for you, give it some name and hit enter. Then connect from mma via MATLink with the same link. Or just Install@LinkConnect@"<name you entered in the terminal>"

Yeah, it is hanging at the executable. Those logs show that it's not progressing past the "launch the binary" phase.

@TheToad I ran ConnectEngine without any errors the first time, but then MEvaluate said "the MATLAB workspace is already closed." Now it acts like there isn't a definition for ConnectEngine and MEvaluate hangs.

(after restart the same thing happened, except MEvaluate does not hang)

Hmm. This is strange. Unfortunately, that's about as far I can do without mma or matlab to test and without getting into low level stuff, which is Szabolcs' expertise. I think you might have to set the paths to the library manually (maybe they're missing?) in mengine.sh or worst case, recompile. You'll need csh (which you have) and g++. I don't know what other changes need to be made to the makefile (in Engine/src/makefile.osx), but @Szabolcs will be able to help you with that.

Sorry 'bout that, but we'll get this working soon :)

@TheToad Alright :) Here's another clue: I just tried it on MMA9 and it worked right away, same system and everything only different version of MMA. MMA 10.0 does not work either.

@Pickett 2014a had a bug that caused this, but 2014b should be fixed. It's late now and I'm tired but I'll try to take a look tomorrow.

@Sektor because the CPU is reasonable for FLOPS/W, but there is no cache, the local memory is tiny, bandwidth to main memory is too little, and there is no way to couple more than one of these devices with a low latency, high bandwidth interconnect. This will make achieving peak performance extremely difficult even for simple, embarrassingly parallel tasks, and scaling efficiently to larger or more complex workloads will be almost impossible. 5 GFLOPS/W sounds great on paper but

a large fraction of useful workloads are ruled out by the fact of being too large to fit into the 32KB local memory, or requiring synchronization

Note that by supercomputer standards Blue Gene/Q has very good energy efficiency: >2 GFLOPS/W. But this is measured LINPACK FLOPS and not some theoretical estimate. It is possible with large numbers of relatively weak CPUs because of an incredible interconnect. At the other end of the scale there are strong CPUs that dissipate 200W but can perform at 1 TFLOP levels. This eases the requirements on the interconnect at least to some extent for moderate scale workloads.