[argobots-discuss] how to marshal/serialize an ABT_pool id for network use

[Phil Carns]

Actually, after thinking through a preliminary design a little more, I 
think I would probably elect to do this mapping myself anyway (even if 
an integer identifier were available in Argobots) so that I can choose 
the size of the identifier in my protocol.  I may need to use a 
relatively small integer due to msg header size constraints.

-Phil

On 02/26/2017 08:46 AM, Phil Carns wrote:
> Hi all,
>
> I'm considering how to build a network service in which incoming 
> requests could potentially be serviced by more than one Argobots pool 
> (ABT_pool).  At the most fundamental level, this will look like a 
> single centralized progress loop for the network that decodes the 
> header for incoming messages, and based on that header calls 
> ABT_thread_create() to launch the handler on a particular pool.
>
> I'd like the client to be able to control which pool it is targeting.  
> I think this means that I should probably create my own data structure 
> that maps "pool_id -> ABT_pool" so that I have a generic identifier 
> that I can share over the network with other hosts.  "pool_id" in this 
> case would just be an int64_t or something similar of a fixed size at 
> any rate.  When a request arrives I would decode the header, look up 
> the pool_id to ABT_pool mapping, then call ABT_thread_create() on that 
> pool.
>
> Before I implement that, though, I just wanted to check- there isn't 
> any existing mechanism built into Argobots for generating stable IDs 
> for a given ABT_pool, is there?
>
> thanks!
> -Phil
>
> _______________________________________________
> discuss mailing list
> discuss at lists.argobots.org
> https://lists.argobots.org/mailman/listinfo/discuss