kilabit.info
Simple | Small | Stable | Secure

Parallel, whatever

SQLite creator, the one who have a second greatest quote

"May you do good and not evil. May you find forgiveness for yourself and forgive others. May you share freely, never taking more than you give."

that I have ever read--at least until now, is the one who point me (indirectly) first about the bad of thread with linked to to Edward A. Lee papers "The Problem With Threads".

From my point of view and by looking current hardware development progress, I think he is right but still I have some problem in my mind.

A small programmer like me usually use 2/4 processor, C, and Linux for having fun (yeah, typical ;) Someone called it a "desktop" user, that write and use program in a little box. The problem is: if thread is bad, what can I do to not use thread, or what is the solution if I face some problem that need to be parallel? Did "thread" was so bad?

So, after reading Lee's paper I got a little thinking about it, and need one answer. I go through Google doing some search about current progress in parallel programming, distributed computing, or whatever you called it.

I search, I read, a lots.

They are a lot of alternative that Lee suggest in their paper, some of them is using some programming pattern/technique/concept, language, and library. The one that I skip here is a language, forget it, like he said, it's a religious thing. Now we got two alternative, pattern and library.

Some of pattern that Lee suggests (and others), is OK. But the problem with pattern is, first, it depend on the problem that you want to solve; second, it need time to fully understand the concept, but when you understand it you get what you pay, and time to implement the concept into code.

The last solution is library. One advantage with library is it use some pattern/concept built into it. There are three library that I found meet with my language requirement: OpenMP, MPI, and Phoenix (MapReduce version for C). First, Phoenix and MPI, last time I check this two library is good if you have a cluster and big data, even if it can be used in one box it like "killing a fly with a bazooka". And I did not want to give a lecture or judging about peoples here, but when you want to write a program especially library or API you must think how others programmer will use it, so Keep It Simple Stupid! is my motto.

OpenMP is a good concept, its designed and developed by an experts, its have a great and clean syntax and module that easy to use by a new programmer (I underline the 'designed' word because I did not known whether the person who design it also participate in developing it too) There is disadvantages on it; First, you need the latest compiler to used OpenMP, gcc implement it since version 4.2.3, but I think that is not a big problem; Second, imagine if there is a little tiny bity mini bug in OpenMP implementation, and your program is core dumped and you have been debugging all day and night and can not found where the bug is. but that what a life is for, right? What a boring life without a problem.

Let just say that my first question is already answered by OpenMP, now go to the next question "did thread is so bad?" Remember when some one say that "goto" is evil? Really? goto is not evil if you know when and where to use it, just like thread. Thread is not evil if you know when to thread, where to thread, and what to thread. But yes, debugging threaded program is really pain in the butt.

I think nobody will ask what is my first greatest quote is, but I give you my third greatest quote,

"If you think I am wrong, then you have right to think of that. If you think you are right, then you have wrong to think of that."

Maybe I am wrong, maybe not. We will see later.