> Perhaps you should use uDAPL/kDAPL ( http://www.datcollaborative.org/ ).
 > It's more documented and it's more portable. It works with Infiniband and 
 > iWARP
 > (http://en.wikipedia.org/wiki/IWARP).

Oh no!!

First of all I don't know of any kDAPL implementation for any OS that
is still being developed.  Everyone completely gave up on kDAPL a long
time ago.

And if all you care about is being able to work on top of IB and
iWARP, then libibverbs + librdmacm works perfectly fine without having
to add another layer and all the complexity of DAPL.  And you don't
have to worry about code like (from dapl/common/dapl_cookie.c):

    new_head = (dapl_os_atomic_read (&buffer->head) + 1) % buffer->pool_size;

    if ( new_head == dapl_os_atomic_read (&buffer->tail) )
    {
        dat_status = DAT_INSUFFICIENT_RESOURCES;
        goto bail;
    }
    else
    {
        dapl_os_atomic_set (&buffer->head, new_head);

        *cookie_ptr = &buffer->pool[dapl_os_atomic_read (&buffer->head)];
        dat_status = DAT_SUCCESS;
    }
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to