[Pvfs2-developers] Distribution by hostname

[Sam Lang]

On Oct 10, 2006, at 2:16 PM, Pete Wyckoff wrote:

> slang at mcs.anl.gov wrote on Tue, 10 Oct 2006 14:01 -0500:
>> On Oct 10, 2006, at 1:02 PM, Pete Wyckoff wrote:
>>> How about just add PVFS_sys_dist_get_IO_server_aliases() and
>>> PVFS_sys_dist_set_IO_server_mapping().  Under the hood each
>>> distribution orders the IO servers it'll use according to the
>>> mapping then does business as usual.  This instead of the random
>>> starting point and sequential ordering used now.
>>
>> What are the benefits of those functions over something like:
>>
>> PVFS_sys_dist_set_param(..., "io-server-mapping", &server_map);
>>
>> I dislike the idea of adding interfaces that won't get much use, esp
>> if we already have a generic mechanism for doing the same.
>
> Good suggestion; I'm fine with this too.  It leaves it up to the
> particular distribution to implement, but they can all call some
> generic functions internally for aspects that are similar.
>
> I'd like it if the type of server_map were some nice thing, maybe a
> "int length" and some const strings that were gotten by asking for
> the list of servers.  It would be good if code can programatically
> pick out the servers it wants to construct a server_map.

Hmm...interesting.  In theory the distribution could even go over the  
network and figure out the best distribution (list of servers) based  
on the parameters specified.  Sort of a load balancing framework...

>
> But doing this through a _dist_set_param call means it's not a hint.
> It's mandatory.  The create that uses this dist should fail if it
> can't manage to apply the requested io-server-mapping.  And this
> makes it part of the distribution API, where it should be.

Right. Inside the create sm I would imagine calling dist- 
 >get_io_server_mapping, and then applying that list to a modified  
version of PINT_cached_config_get_next_io that would mostly just do  
checking of the list (or returning all the servers as it does now if  
no list were provided).

>
> 		-- Pete
>