PolyORB/Examples/Eratho

This is a distributed Prime Finder program. 

Each partition is in charge of testing whether a given number is
divisible by its local prime. If not, then a RPC is performed and the
next task running on the next partition is asked to perform the same
divisibility test with the next prime number. If there is no more
prime to test, then the current number is a new prime. A new cell is
allocated on the partition and the current number is added to the
local prime table of this partition.

When numbers between 2 .. 50 have been tested, partitions terminate
and a distributed termination detection is activated.

There are four examples of the same program.

    spiral (Plain RPC) --- (+ Object oriented) ---> dynamic

       |                                               |
(+ Asynchronous)                                (+ Asynchronous) 
       |                                               |
       v                                               v

    cycle              --- (+ Object oriented) ---> absolute

To build this program, once in the "Examples/Eratho" directory, type
"make". To execute the distributed program type "./mainloop".


spiral:

   Prime_1, Prime_2 and Prime_3 are three identical RCI packages
   (sorry, generic RCI packages are not yet implemented). They are
   connected together into a token ring where the token is the number
   to test. When procedure Test_Primarity is invoked, a RCI Prime
   package tests if its current prime divides the number. If not, a
   remote procedure call is invoked inside the current remote
   procedure call. The successive remote calls return the
   divider. This is an example of recursive remote procedure calls.
   Note that RCI packages Prime_1, Prime_2 and Prime_3 can't be the
   same RCI package assigned three times, because a remote call
   interface library unit shall be assigned to at most one partition
   of a given program.

cycle:

   Prime_1, Prime_2 and Prime_3 are still three identical RCI
   packages. But rather invoking a remote procedure call inside a
   remote procedure call, procedure Test_Primarity invokes an
   asynchronous remote procedure call and doesn't keep waiting for the
   returned parameter Divider of example "spiral". In this case, when
   the algorithm ends up, the last invoked partition executes another
   asynchronous remote procedure call to release a semaphore and to
   provide to the main partition the result.
   
dynamic :

   There are several instances of the same partition (well, these
   instances have different names, but they contain the same code).
   In partition2 and partition3, a normal package Prime derives
   New_Pool_Type from Prime_Pool_Type. Note that on each of these
   partitions, the derivation is different, because Prime is a
   (replicated) normal package. A reference to an object declared in
   each of these partitions is registered to Controller. When all
   these references are registered (3), Controller builds a token ring
   and provides to each partition a reference to the object of its
   neighbor. Thus, we have the same behaviour as spiral's one, except
   that partitions communicate through remote dispatching procedure
   calls.

absolute:

   This example is a merge of "cycle" and "dynamic" examples.


The "dynamic" and "absolute" examples provide a good example of way to
implement distributed applications based on the client and server
model. For instance, in example "dynamic", partition2 and partition3
are identical. We can imagine to have the following configuration file :

configuration Dynamic is
   pragma Starter (None);
   Partition1 : Partition := (Controller);
   Partition2 : Partition := (Prime);
   for Partition'Storage_Dir use "bin";
   procedure Mainloop is in Partition1;
end Dynamic;

And to start independently :

host1 % bin/partition1 --boot_server tcp://host1:unused-port --nolaunch

And to start several instances of partition2 :

host2 % bin/partition2 --boot_server tcp://host1:unused-port --slave

host3 % bin/partition2 --boot_server tcp://host1:unused-port --slave

host4 % bin/partition2 --boot_server tcp://host1:unused-port --slave

In our example, we need 3 instances of prime, because the prime finder
algorithm starts running when 3 clients have been registered.

This is an example of client / server applications using advanced
features of Ada 95 Distributed System Annex (ie distributed objects or
RACW types). Of course, you can also write client / server
applications by using plain Ada remote procedure calls. In this
context, several instances of the same partition could perform Ada
remote procedure calls to the same partition as long as these clients
don't include RCI packages (See E.2.3 (17)).