Implementing shared/exclusive locking (readers/writer) lock
Question Implementing shared/exclusive locking (readers/writer) lock
Hi,
I am using semaphore to implement Single Writer Multiple Readers (Reader / writer) lock.
The problem is that semaphores are critical resources in linux systems, and there is a limit on the number of semaphores
that you can create.
So, after creating some semaphores, sometimes my application exceeds this limit, and I get an error like 'No space left on
device.'.
I have to reboot the system sometimes, to release the semaphores, if my application is terminated abruptly!
How should I implement readers/writers lock using something else than semaphore, which is scalable?
I don't need reader/writer locks across multiple processes. I just need to use them in multiple threads of my application,
which is a single process.
But I want a very scalable solution, say I want 500 locks through out my application daemon's lifetime.
I read in the link below that postgres also suffered the same problem, but there it is recommended that the limit should be
increased, which I don't want users to do.
http://www2.units.it/~nircdc/doc/postgres/kernel-resources.html
Sabyasachi.
Hi,
I am using semaphore to implement Single Writer Multiple Readers (Reader / writer) lock.
The problem is that semaphores are critical resources in linux systems, and there is a limit on the number of semaphores
that you can create.
So, after creating some semaphores, sometimes my application exceeds this limit, and I get an error like 'No space left on
device.'.
I have to reboot the system sometimes, to release the semaphores, if my application is terminated abruptly!
How should I implement readers/writers lock using something else than semaphore, which is scalable?
I don't need reader/writer locks across multiple processes. I just need to use them in multiple threads of my application,
which is a single process.
But I want a very scalable solution, say I want 500 locks through out my application daemon's lifetime.
I read in the link below that postgres also suffered the same problem, but there it is recommended that the limit should be
increased, which I don't want users to do.
http://www2.units.it/~nircdc/doc/postgres/kernel-resources.html
Sabyasachi.
Slightly non-portable are pthread read-write locks (man pthread_rwlock_init) as it requires you to #define __USE_UNIX98.
Alternatively, you could use a pthread_mutex_t (man pthread_mutex_init). You'll need a couple of them plus some ancillary data to create your own reader/writer lock using pthread_mutexes.
Also, you might want to man 8 ipcrm ... ipcrm is a command-line tool that allows you to delete kernel semaphores without having to reboot.
Alternatively, you could use a pthread_mutex_t (man pthread_mutex_init). You'll need a couple of them plus some ancillary data to create your own reader/writer lock using pthread_mutexes.
Also, you might want to man 8 ipcrm ... ipcrm is a command-line tool that allows you to delete kernel semaphores without having to reboot.
Wouldn't you benefit from having more concise exception handling to ensure if something does go wrong, all of your semaphores are released?
The following threads are about preventing multiple instances of an application from coexisting, but the concepts are the same and should prove you useful:
http://www.linuxquestions.org/questions/programming-9/creating-single-instance-linux-application-256693/
http://www.linuxquestions.org/questions/programming-9/restricting-multiple-instance-of-a-program-242069/
Hope this helps.
http://www.linuxquestions.org/questions/programming-9/creating-single-instance-linux-application-256693/
http://www.linuxquestions.org/questions/programming-9/restricting-multiple-instance-of-a-program-242069/
Hope this helps.
Topic archived. No new replies allowed.