linux - How to prevent writer starvation in a read write lock in pthreads -


I have some questions about reading POSIX pthreads on a * nix system-write locks, for example, says Linux is.

I want to know what is the default bias for the lock to read, likes to write it or vice versa? Does it provide some APIs to replace this default behavior?

Does the posix pthread provide some APIs so that we can replace pthread_rwlock_t to stop the author's hunger? What I have read (if I am wrong, correct me), the default implementation is biased towards reader threads and therefore the author threads can face starvation.

I have read sample implementations of RW Lock from this book, programming with Pozzys Threads by David Boothhoff.

I want to know how to handle the ground water of author Thread? Are some APIs used that can be used to set the properties of the lock locks which prevent them from writing starvation (I have never heard of it)? Or does the user handle this problem?

If you think the answer is implementation-defining, please give me an example of how it is in Linux, because what I am looking for.

Please note that I only want a * nix system solution Do not think that I am rude, but posting a few windows-specific codes is useless for me.

Thanks for all your help and patience :)

it really depends on implementation - so when you ask specifically about Linux, then refer to current Anpitiel implementation of my comments pthreads, which is used in modern Glibik.

Here are two related, but different, here are the problems. First of all, this is the situation:

  • There are current locally held locks, and the authors are waiting. A new thread tries to take a reading lock.

The default action here allows the reader to proceed - effectively "climbing the queue" on the author. However, you can override it. If you use the are pthread_rwlockattr_setkind_np () function to install PTHREAD_RWLOCK_PREFER_WRITER_NONRECURSIVE_NP flag attr that you have pthread_rwlock_init () < / Code>, your call will block the reader in the above position.

The second position is:

  • The last holder issues a lock, and both readers and authors are waiting.

In this situation, NPTL will always wake the reader to a writer in priority.

taken together, the above means that if you PTHREAD_RWLOCK_PREFER_WRITER_NONRECURSIVE_NP flag, its authors should not be hungry (of course, now hungry for a steady stream of readers of the authors compete c'est ). All of you can confirm this by checking the sources in you (this is all very readable).

Note that there is also a PTHREAD_RWLOCK_PREFER_WRITER_NP , but it looks like not to the right impression - quite possibly a bug ( or possibly No - see ).


Comments

Post a Comment

Popular posts from this blog

sql - dynamically varied number of conditions in the 'where' statement using LINQ -

I am using my first project LINQ (in mvc), so maybe there's something very simple that I remember. However, a day of searching and experimenting does not work, so it works. I am trying to write a LINQ query (Linux to SQL) in which in many cases where the statement is different from an OR or an AND. We do not know how many conditions are going in the query till runtime. This search is for filter control, where the user can choose multiple criteria for filtering. Select from table where table.col = 1 or table.col = 2 or table.col = 7 .... 'Other number of conditions Before I just create the SQL query as a string, while looping over all the situations. However, it seems that there should be a good way to do this in LINQ. I tried to use Expression trees, but they had a second thought for this moment on my head, where the lambda function was to be executed inside the statement such as: values ​​for matchingRows = matchingRows for each value Where (function (row) row.c...
Read more

asp.net mvc - Dynamically Generated Ajax.BeginForm -

I have a problem with ASP.NET MVC Ajax forms. I generate many Ajax form in a table such as: & Lt; Input type = "submit" name = "activity" value = "add activity" / & gt; & Lt;%}% & gt; & Lt; / TD & gt; & Lt; / TR & gt; & Lt;%}% & gt; So far so good. When I post one of those forms, it gives a partial view in a div, and in this new partial view a new Ajax Beginnform (which is not in Genesis form) This second Ajax Beginnorf is exactly like the ones I got on the table, but when I post it, I do not get the "Ajax Post", but a normal post that sends me to a new page, I do not know whether the two Ajax forms What is the reason for the difference between, why is it the "dynamically generated" Ajax form behaves like a normal HTML form? By the way, I know that I can do it in different ways using javascript, but I would like to understand whether it can only be done using MVC helplars and Ajax lib...
Read more

Debug on symbian -

I am using trk for phone debug This works well for the Holorld project But when I start the project in phone debug mode 1) the load failed 2) TrkProtocolPlugin: to target the specified file Failure (please verify If any body understands what problem I am experiencing, then this problem helps me Pre-release If the application has the privileges assigned (with the appropriate certificate) itemprop = "text" > In your case, I will check: / li> If there is no conflict of application ID with any other application of device If there are problematic commands in the installation package (for example copy copy for non-accessible directories)
Read more