An arbitrary number of aliases is useful in a number of ways.
For example you mention wanting to register an OPAL MCA parameter and
later alias it as an OMPI MCA parameter. What if we also wanted to
alias it as an ORTE level parameter. The best example I can think of
is the TCP include/exclude interface MCA paramters. OPAL may need to
know them for the 'if' functionality, ORTE for the OOB/tcp component,
and OMPI for the BTL/tcp component. Another example might be the
TMP_DIR MCA parameter discussion that has been going on in another
thread. Why not have an OPAL/ORTE/OMPI variant of this parameter?
Another reason for multiple aliases is general code development. Going
forward we may want to re-alias an already aliased MCA parameter to
give it a 'better' name.
On Apr 2, 2008, at 10:50 AM, Sharon Melamed wrote:
> -----Original Message-----
> From: devel-core-bounces_at_[hidden]
> [mailto:devel-core-bounces_at_[hidden]] On Behalf Of Jeff Squyres
> Sent: Wednesday, April 02, 2008 3:44 PM
> To: Open MPI Core Developers
> Subject: Re: [devel-core] [RFC] Add an alias name to MCA parameter
> BTW, these mails can go across devel.
> devel-core is only for "private" stuff, like dialup phone numbers,
> On Apr 2, 2008, at 9:34 AM, Jeff Squyres wrote:
>> I agree that it would be beneficial to support an arbitrary number of
>> Also, some points about ompi_info:
>> - say you register "opal_paffinity_base_alone", and later register
>> "mpi_paffinity_alone" as an alias. When you "ompi_info --param mpi
>> all", the alias should still show up
>> - aliases that are displayed through ompi_info's --param option
>> somehow indicate that they are aliases, and show the "real" name as
>> well. Perhaps something like this? (just an idea)
>> MCA btl: parameter "mpi_paffinity_alone" (current
>> value: 1,
>> alias for: opal_paffinity_base_alone)
>> If nonzero, assume that this job is the
>> only (set of)
>> process(es) running on each node and bind
>> processes to
>> processors, starting with processor ID 0
>> On Apr 2, 2008, at 8:37 AM, Josh Hursey wrote:
>>> This sounds great. I have a couple questions though:
>>> - Is there a patch for this that we can look at/test?
>>> - Do you require that the parameter be registered before adding an
>>> - Is the 'index' argument referencing the original MCA parameter, or
>>> are aliases given individual 'index' values?
>>> - Does this support more than one alias for a single MCA parameter?
>>> If so then there should be a way to specify that in the remove
>>> On Apr 2, 2008, at 9:22 AM, Sharon Melamed wrote:
>>>> WHAT: Add an alias name to MCA parameter.
>>>> WHY: There is a parameter that we need to register and use in OPAL
>>>> (before ompi_init) but historically the parameter name is
>>>> ompi_something_.... With the alias name we can register this
>>>> parameter in OPAL and call it opal_something_.... and then add an
>>>> alias name: ompi_something.. Now the user can use this parameter
>>>> with its real name or with its alias.
>>>> WHERE: in /opal/mca/base/
>>>> TIMOUT: Thursday - April, 10.
>>>> Add two Interfaces to the MCA system:
>>>> OPAL_DECLSPEC int mca_base_param_add_alias (int index, const char*
>>>> OPAL_DECLSPEC int mca_base_param_remove_alias (int index);
>>>> These functions could be called any where in the code after the
>>>> registration of the MCA parameter. (mca_base_register)
>>>> This change includes:
>>>> . Adding a member to mca_base_param_t structure.
>>>> . Modifying the find functions
>>>> . Modifying ompi_info.
>>>> devel-core mailing list
>>> devel-core mailing list
>> Jeff Squyres
>> Cisco Systems
>> devel-core mailing list
> Jeff Squyres
> Cisco Systems
> devel-core mailing list
> devel mailing list