[openib-general] Minor question on fat tree routing

[Yevgeny Kliteynik]

Hal Rosenstock wrote:
> On Wed, 2006-12-20 at 10:02, Yevgeny Kliteynik wrote:
>> Hi Hal,
>>
>> Hal Rosenstock wrote:
>>> Hi Yevgeny,
>>>
>>> Minor question on fat tree routing:
>>>
>>> osm_ucast_ftree.c:__osm_ftree_construct_fabric has the following code:
>>>
>>>    if (__osm_ftree_fabric_perform_ranking(p_ftree) != 0)
>>>    {
>>>       if (__osm_ftree_fabric_get_rank(p_ftree) > FAT_TREE_MAX_RANK)
>>>
>>> Should < FAT_TREE_MIN_RANK also be checked there too ? Does it fallback
>>> to default routing for this case too ?
>> This is also checked, but as part of more earlier checks in the same function:
>> FatTree routing will abort even before ranking the tree and fallback to the default
>> routing if a fabric has less than 2 switches.
> 
> What about 2 or more switches but rank is 1 ? Isn't that possible too ?

2 or more switches and tree rank 1 means that all the switches are leaf switches,
which means that they all connected directly to HCAs.
So either these switches are not connected to each other, which means that we
actually have several disconnected subnets, or they are connected to each other,
which means that they have connections the same rank of the tree, which is illegal 
and is discovered by indexing.

But I agree - adding the (< FAT_TREE_MIN_RANK) check will improve readability.

-- Yevgeny
 
> -- Hal
> 
>> -- Yevgeny
>>> Thanks.
>>>
>>> -- Hal
>>>
>>>
>>> _______________________________________________
>>> openib-general mailing list
>>> openib-general at openib.org
>>> http://openib.org/mailman/listinfo/openib-general
>>>
>>> To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
>>>
>