Re: When to use forks, when to use threads ...?

[wojtyk]

Well, they each certainly have their pros and cons. Personally, I opt for forking unless I have something that requires a significant amount of shared interaction (either through required shared memory or producer/consumer queues). I find the extra overhead of forking to be minimal in modern day operating systems, and the gains are significant:

- fewer memory leaks
- process isolation
- fewer race conditions
- deadlocks/starvation not an issue

Look to the latest release of Google's Chrome as a perfect example. The current threaded asynchronous model of browsers isn't cutting it, so they're moving to a process-per-tab model for all of the above reasons.

Of course, this is a personal opinion and therefore flame-worthy, so take all advice with grain of salt :)

[mr_mischief]

It is possible to deadlock two separate processes that communicate with one another if the protocol between them is not well designed. The "fewer memory leaks" claim isn't necessarily true either, although it's easier to clean up after a small leak because the OS will reclaim the memory once the process exits.

[wojtyk]

It is possible to deadlock two separate processes that communicate with one another if the protocol between them is not well designed

Fair enough, but still seems like a nitpick (in that it is possible, but not probable - whereas in the case of programming threaded apps, it's a common occurrence). Also in the case of separate processes, it's much easier to recognize & debug.

The "fewer memory leaks" claim isn't necessarily true either

This one I stand by (in the statistical average case), particularly as the software grows more complex. As you mentioned, process termination is a natural "reset switch" that does wonders in keeping leaks down. But there's a whole host of other reasons that make memory leaks more common in threaded apps:

- Harder to recognize and pinpoint (and thus fix...), since "memory usage" reported will be a macro-level observation instead of a process-isolated observation
- More frequent use of resource locks (ie semaphores), reference counting, and shared resources in threaded apps, making it easier to leave resources in a limbo state
- Maybe a redundant point, but "increased coding complexity" when using threads increases the possibility of a coding error


As with all things involving coding, _any_ piece of code could be done "right" with no memory leaks, but I believe it's much easier to both introduce leaks AND harder to pinpoint and fix them in threaded apps vs standalone processes.

[mr_mischief]

I agree in general, although I don't have enough data to analyze anything statistically. I'm just pointing out that "less likely" is not the same as "cannot happen" and that "probably fewer" doesn't mean "definitely fewer". Your earlier post seemed absolute. The same programmer writing the same project in the two different methods seems likely to experience fewer leak and deadlock problems with processes than with threads. That's not necessarily the case, though. "Necessarily" is the key word.