GIOS M9: Inter-Process Communication

What is Inter-Process Communication?

Inter-Process Communication (IPC) refers to OS-supported mechanisms for interaction (coordination + communication) among processes. IPC methods are loosely grouped into two main categories:

• message-based IPC: sockets, pipes, message queues.
• memory-based IPC: shared memory, memory mapped files.

The end result of any IPC mechanism is that data is transferred from one address space into another. As part of message-based IPC mechanisms, the CPU must copy data to/from a port; as part of memory-based mechanisms, the CPU must define VA-PA mappings. If the data is large, the CPU overhead required to copy data is much larger than that to map memory.

Message-Based IPC

Message passing involves the send/recv of messages. The OS handles channel management + synchronization, enabling one process to write to a port and the other to read from it. There are many different POSIX-supported implementations of message-based IPC:

• Pipes are characterized by two endpoints, and carry a byte stream between two processes. Pipes are commonly used to connect the output of one process to the input of another.
• Message Queues are organized such that the sending process sends a properly formatted message to a channel, and the receiving process receives a properly formatted response. The OS may handle things like message parsing, formatting, prioritization, and scheduling.
• Sockets are used to directly send and recv messages via the socket API. Unlike the previous two mechanisms, processes can be used for IPC across different machines.

Memory-Based IPC

In Shared Memory IPC, processes read and write to the same shared memory buffer. P1 and P2 will use different virtual addresses, but map to the same physical addresses in RAM. System calls are only used for the setup process; once the OS establishes the shared memory buffer / VA-PA mappings, it is no longer needed. However, processes must use synchronization to access shared memory region.

Shared Memory

Specific Implementations

SysV is the UNIX standard for shared memory. As part of SysV, the OS provides shared memory in the form of segments - dynamically-sized and not-necessarily-contiguous physical pages.

Using shared memory as part of SysV typically works as follows:

1. Create $\rightarrow$ OS allocates shared memory and assigns it a unique ID. Process(es) provides a key to the OS in return for the ID. ftok(pathname, proj_id) (hash function to generate key) shmget(key, size, flag)
2. Map $\rightarrow$ attach shared physical memory to each virtual address space. shmat(shmid, addr, flags)
3. Detach $\rightarrow$ invalidate address mappings to prevent virtual addresses from accessing. Segment still exists in physical memory, and may be re-attached (to same or different processes) as desired. shmdt(shmaddr)
4. Destroy $\rightarrow$ explicitly deallocate physical memory corresponding to segment. shmctl(shmid, cm=IPC_RMID, buf)

The POSIX Shared Memory API provides shared memory in the form of files as opposed to segments. It is not as widely supported as SysV. The POSIX shared memory API has analogous operations to SysV:

1. shm_open() $\rightarrow$ returns file descriptor, where file is located in tmpfs file system.
   • Must then use ftruncate(fd, size) to define the size of the shared memory.
2. mmap() and munmap() $\rightarrow$ attaches shared memory to virtual address space.
3. close() $\rightarrow$ removes file descriptor from virtual address space.
4. shm_unlink() $\rightarrow$ indicates to the OS that all shared memory data structures should be deleted + freed.

Synchronization

When data is placed in shared memory, it can be concurrently accessed by all processes which have access to the shared memory. It is therefore important to use Synchronization, as in the case of any concurrent access to a shared resource.

We can manage synchronization of IPC shared memory in a few ways:

• Apply mechanisms supported by process threading library (ex: pthreads).
• Utilize OS-supported IPC for synchronization.

Either method must coordinate concurrent accesses to the shared segment (via mutexes), and indicate when data is available / ready for consumption (condition variables).

PThreads

In order to use pthreads structure as part of IPC, we must ensure that all interacting processes have the same access to IPC data structures. This is not a factor when considering threads, since the structures are by-default available to all threads of the same process.

typedef struct {
	pthread_mutex_t mutex; 
	char *data;
} shm_data_struct, *shm_data_struct_t; 

// create shm segment 
seg = shmget(ftok(arg[0], 120), 1024, IPC_CREAT|IPC_EXCL));
shm_address = shmat(seg, (void *) 0, 0); 
shm_ptr = (shm_data_struct_t)shm_address; 

// create + init mutex 
pthread_mutexattr_setpshared(&m_attr, PTHREAD_PROCESS_SHARED); 
pthread_mutex_init(&shm_prt->mutex, &m_attr); 

---

(all images obtained from Georgia Tech GIOS course materials)