Re: [Caml-list] STM support in OCaml

[Andrae Muys <andrae@internode.on.net>]

On 09/03/2006, at 8:10 AM, Gerd Stolpmann wrote:

> Am Donnerstag, den 09.03.2006, 09:06 +1100 schrieb skaller:
>> On Wed, 2006-03-08 at 14:45 -0600, Brian Hurt wrote:
>>
>>> One comment I will make is that a mutex is expensive, but not *that*
>>> expensive.  I just wrote a quick program (available if anyone  
>>> cares) in
>>> GNU C that measures the cost, in clocks, of locking and unlocking  
>>> a posix
>>> mutex.  On my desktop box (AMD Athlon XP 2200+ 1.8GHz), I'm  
>>> getting a cost
>>> of like 44 clock cycles.  Which makes it less expensive than an  
>>> L2 cache
>>> miss.
>
>> I have no idea if Linux, for example, running SMP kernel,
>> is smart enough to know if a mutex is shared between two
>> processing units or not: AFAIK Linux doesn't support
>> interprocess mutex. Windows does. Be interesting to
>> compare.
>
> Of course POSIX supports interprocess mutexes: Mutexes are inherited
> across fork().

As skeller suggested, I can confirm this does not work.  Mutexes are  
not OS resources, they are initialised regions of memory (generally a  
word).  As fork() does copy-on-write, this leads to the mutex being  
duplicated NOT shared; and yes given that fork() only duplicates the  
calling thread, that does mean you can end-up with mutexes locked in  
the child with no  controlling thread available to unlock them.

fork() safety is an important issue to be aware of when working with  
posix-threads.  The best reference I am familiar with is Programming  
With Threads by Kleiman, Shah, and Smaalders, which includes a whole  
chapter on the this and related topics.

> And Linux even implements threads as processes, so you
> can even place mutexes in shared memory (but do not ask me how).

I've done this in an embedded app I wrote a few years back.  It's not  
hard, you create the shared-memory segment, cast some of it to be of  
type pthread_mutex_t and pass it to pthread_mutex_init()  (I'm  
working from memory here, so the function/type names may be off).

> Anyway, measuring the costs of a mutex is probably not simple. They  
> are
> highly optimized for the frequently occurring cases. And today "SMP"
> behaves often more like NUMA, especially for multi-core CPUs.

Agreed, however in my experience, if optimistic locking isn't going  
to work for you then neither are mutex-based shared-memory approaches.

Andrae Muys

-- 
Andrae Muys
andrae@netymon.com
Principal Kowari Consultant
Netymon Pty Ltd