[Pvfs2-developers] Distribution by hostname

[Sam Lang]

On Oct 10, 2006, at 1:02 PM, Pete Wyckoff wrote:

> rross at mcs.anl.gov wrote on Mon, 09 Oct 2006 10:57 -0500:
>> I'm warming to the idea of doing everything related to distributions
>> through the hints mechanism, even though it sort of undoes Pete's
>> attempt to simplify the discussion.
>
> My number one complaint with using a hint mechanism is that it is a
> direct indication that the API was poorly designed in the first
> place.
>
> An exception to this rule is to implement purely optional parameters
> for performance optimization, like most of the hints applied to
> MPI_File_open.  The code still works without these hints, just
> slower.
>
> Like Walt, I'm also not real clear on why anybody needs to specify
> hostnames as well as a distribution when creating a file.  But the
> discussion I've seen makes it look like people who want to do this
> do not view the hostname placement rules as optional.  If PVFS
> ignores the hint, that is definitely not okay.  So it should be part
> of the API proper if we take it seriously.
>
> I don't care much about passing strings versus passing arrays of
> keyvals, but I don't want to have a whole set of different knobs
> to control what are conceptually part of the same function: how do
> the bytes of the file get laid out on disks.
>
>> So I think we're mostly trying to work out what our API should really
>> be, whether we should extend the distro functionality vs. going  
>> totally
>> to hints, and if we go to hints what that API should look like,  
>> right?
>
> Yes.  Pick one.  If you want to augment the existing PVFS_sys_dist
> set of calls (lookup, setparam, ...) to handle hostnames, fine.  If
> you want to scrap all that and go with a big opaque string passed
> into PVFS_sys_create, that would be another approach, but let's not
> do both.  (And not call the string a "hint" if it is mandatory.)
>
> 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.

-sam


>
> 		-- Pete
>
> _______________________________________________
> Pvfs2-developers mailing list
> Pvfs2-developers at beowulf-underground.org
> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
>