Tangled Threads

[billroper]

I've been doing some experimental work to see if I can multithread our calculations. This involved a lot of code refactoring. I am now ready to see if the refactored code survives running with one thread. If it does, then I'll try two so I can watch it crash...

What? You think these things work the first time?

Comments

[mia-mcdavid.livejournal.com]

No, the question is not, *will* it crash, but *where* will it crash. This time.

[giza.livejournal.com]

Threading is evil!

*runs and hides*

[madfilkentist]

Trouble with threads is, your test will run fine, and the code will wait till it's in production to find a race condition and crash.

[unkbar.livejournal.com]

Are you running on multiprocesor or multicore systems? Multithreading won't do much to speed up large blocks of calculations on a uniprocessor. In a uniprocessor, splitting a big calculation in two and running it in two concurrent threads doesn't give you any more MIPS, and may add thread management overhead. Threading's big win is things like keeping the GUI live while a long calculation is going on, or handling network traffic concurrently with file I/O.

[billroper]

Oh, yes, I'm fully aware that multiprocessing would simply waste cycles on single processor, single core systems. But on the servers, we have multiple processors and I'm expecting a lot of multicore desktops in the near future.

Since our calculations are dependency driven, it's not too hard to multithread them. Mostly. :)

[kevinnickerson.livejournal.com]

I've had cow-orkers in the past that thought they worked first time. Even to the point of checking in code that didn't compile, or forcing a corrupted zip of code that didn't even work to Austrailia.