Hi,

Lets say, I have a reader-writer system where reader and writer are concurrently running. 'a' and 'b' are two shared variables, which are related to each other, so modification to them needs to be an atomic operation.

A reader-writer system can be of the following types:

1. rr
2. ww
3. r-w
4. r-ww
5. rr-w
6. rr-ww

where
[ r : single reader
rr: multiple reader
w : single writer
ww: multiple writer ]

Now, We can have a read method for a reader and a write method for a writer as follows. I have written them system type wise.

1. rr

read_method
{ read a; read b; }

2. ww

write_method
{ lock(m); write a; write b; unlock(m); }

3. r-w
4. r-ww
5. rr-w
6. rr-ww

read_method
{ lock(m); read a; read b; unlock(m); }

write_method
{ lock(m); write a; write b; unlock(m); }


For multiple reader system, shared variable access doesn't need to be atomic.

For multiple writer system, shared variable access need to be atomic, so locked with 'm'.

But, for system types 3 to 6, is my read_method and write_method correct? How can I improve?

Sincerely,
Srinivas Nayak

The only way to improve on this is to use read/write locks. They're similar to mutexes, except that there are two locking methods - a way to read_lock them and a way to write_lock them. Any number of threads can get a read_lock, but if a thread requests a write_lock, no other threads can have any type of lock on it.

Have a look at the man pages for pthread_rwlock_* (start with pthread_rwlock_init()) if you're using pthreads, but a lot of higher-level threading libraries provide their own implementation.

John G

Hi -

Let's be clear:

1. You've got concurrency, or you don't.

2. If a global variable is being accessed concurrently, you ALWAYS need locking.
3. The distinction is this:
Take a look at Beej's Guide to IPC. It's not about pthreads, but it gives some great times on concurrency issues:

http://beej.us/guide/bgipc/

'Hope that helps .. PSM