Thread: shmget, can this be "segmented"

Basically, my question is, I need to have multiple clients write requests to a shared memory block and a single server to process these requests and return them to the right owner.

Handling the client placing requests is easy as all clients can just use a lock to put their request in the shared memory.

The question is, how do I allocate a segment of the shared memory to each client, so that the client can read on this segment, and wait for the "server" to place the result in that chunk?

I haven't seen anyway to segment shared memory like this.. perhaps if someone knows some standard functions to point me in the right direction?

Originally Posted by Codeplug

Instead of trying to manage multiple shared blocks (at the OS level), create a single shared block that's able to accommodate up to N simultaneous clients. When a client issues a request to the server, it's allocated a client-response handle/entry somewhere in the shared block. When the client receives the response, it can return the handle/entry back to the server for reuse, or use it again in subsequent requests.

Or you could consider other IPC methods: sockets, pipes, message queue's

gg

Thanks for the response.

It's a course project, so I have to use RPC.

I would like to use a single shared block, but I'm not sure how each client gets allocated a portion of the shared memory in which to write... ideally, if there are 4 clients, each client should be able to write to his portion of the shared memory without interference...

Originally Posted by Codeplug

>> but I'm not sure how each client gets allocated a portion of the shared memory
Simple approach is to just use an array of per-client structures, and an index representing "the next structure to allocate to a client". And once a structure has been given to a client, it's never returned. Then to support N clients, the array would contain N elements. First client gets index 0, next client gets index 1, etc...

If you need to support the notion re-usable client structures, then you would need to scan the array for the next available structure to assign.

gg

I see. This is starting to make a lot of sense now.

so I would need one general holder for each client to "register" himself. The server reads from this place, and then returns a pointer to some chunk of the shared memory to the new client. all interactions after that go between those new chunks...