I was reading the difference between POSIX and System V semaphores, and I read some articles on the same. In each article this statement is written : "System V semaphores are beneficial if you need to implement atomic operations with multiple increments-decrements in a single step."
My questions are:
1) What is the need for multiple increment/decrement in a single atomic operation? Can you please explain with an example?
2) Why does semop allow the value to be changed by a value less than -1 and greater than +1? Is there any practical usage/example of the same?
(I know that with the semop() function in System V semaphores I can have the semaphore increase or decrease by a specified value for than one semaphore in the semaphore array & that the same cannot be done with sem_wait() or sem_post() of POSIX semaphores. But what is the use of the same?)
The articles I read, for your reference:
1) http://www.ibm.com/developerworks/library/l-semaphore/
2) Differences between System V and Posix semaphores
3) http://www.linuxdevcenter.com/pub/a/linux/2007/05/24/semaphores-in-linux.html?page=4
4) http://linuxtips.pbworks.com/w/page/29023300/SystemV%20vs%20Posix%20IPC
UPDATE:
I have read the below articles which have some mention on multiple increment/decrement of a semaphore using semop(), but I still not able to get an example/practical usage of the same. Articles for your reference:
1) http://kaharris.org/teaching/51081/Assignments/Final/systemV.pdf
An excerpt from the article: (under the title "Example of Multiple Semaphore Operations")
"The power of System V semaphores is that they can be used to atomically check and set multiple semaphores in one operation."
He has given a small snippet of how to do it also. But no practical usage of the same.
2) http://www.anirudhtom.com/2011/02/system-v-semaphores-for-babies.html
The author has written a code for multiple increment/decrement of semaphore in an atomic operation under the title "IMPLEMENTING A SET OF SEMAPHORE". The practical usage of the same is still not mentioned here as well.
An excerpt from the book:
"In UNIX System V, the semaphore mechanism makes some adjustment. The operation atomicity is retained. However the added or subtracted value of operations can be greater than one. (Why? What use?). And even more, processes can do multiple semaphore operations simultaneously to avoid deadlock problems when several processes compete for a number of different resources simultaneously. (How? Example?)"
I hope that makes the question more clear.
Please comment if you have any doubts on the question asked.
Thanks in advance!
Example where multiple increment/decrement may be handy: Fancy some kind of conference system where a video session needs one channel for video and the other for sound, but other sessions only use the audio channel. To control access to the channels, semaphores can be used. When a session terminates, it should decrement all semaphores it uses, at once, to free them for other sessions. If this isn't atomic, the video conference session would perhaps, when it starts, be able to grab the audio channel but be too late to also grab the video channel.
Your second question is related to load balancing. By allowing other values than [-1..1], you can use the semaphore value as a queue size meter. In the conference example above, if the load on all local channels are very heavy, perhaps you can redirect your session to another, less busy trunk with its own channels.
Hope this makes it a bit clearer.