What is the difference between shared memory object (created using shm_open) and shared memory segment(shmget)?
Do they have any limitations that cant be tuned like in case of shared memory shmmax and shmall?
Are there any performance variations between these two, and in what IPC scenarios should these be used?
"shmget" is a Linux-Specific method of allocating shared memory implemented in the Linux kernel.
"shm_open" is a library function that emulates shared memory by mapping files using mmap. Because the files are mapped using the "shared" flag the memory is shared between the processes.
In Linux 1.x the "/dev/shm" directory (containing the files) was just a regular directory so the shared memory using "shm_open" really were disk files. In Linux 3.x "/dev/shm" is a special directory to avoid that shared memory really has to be written to disk.
I think both of the two methods can be replaced by the other one. It is only because of historical reasons why there are two different methods to create shared memory.
I found in testing that the memory accessed via shm_open to be faster than the memory accessed via shmget. Other than that, they are pretty similar in terms of features. There must be some subtle cache or TLB differences between the two, but I am not familiar with the details under the hood.
Note I had to use the POPULATE option for shm_open to get this performance increase.
