[nemo] Re: globalHAHA can not guarantee shorter path benefits; can't predict placement of CN

[Alexandru Petrescu <[e-mail]>]

Pascal Thubert (pthubert) wrote:
>> 5.1  Initial routing
>>
>> 5.1.1  External routing
>>
>>    Sites are expected to be connected locally to the internet, via the
>>    network of one or more service provider.  Each site has a default
>>    route to the internet via that connection.
>>
>>                                     . ..   /
>>             ---------          ........ ..;.           ---------
>>            |         |       ..          / .....      |         |
>>            |     ::/0 ->  ....          ;      ...  <- ::/0     |
>>            |         ============HAHA=TUNNEL===========         |
>>            |         | ....           ;          .... |         |
>>            |         | .<- Home::/32 /  Home::/32 ->..|         |
>>             --------- ...           ;             ...  ---------
>>                         .....      /               ..
>>                                .  ;....      ......
>>                                  /  ..........
>>
>>
>>    In return, each site advertises the aggregation that encompasses
> the
>>    Distributed Home Network aggregation back into the service provider
>>    network, to the outside world (the internet).
>>
>> Please note that we have sites and providers here and the prefix
>> announced belongs to the SITE and NOT to the provider. This i read as
>> PI for end sites being injected in the DFZ.
>>
> [Pascal] I did not have time yet to change the text since you raised the
> point. I'll have to word that that the distributed home network can be
> shared among ISPs, and maybe place a shorter prefix in the examples.
> That's a good comment to improve the draft, but that does not change the
> model at all.

Pascal, do you agree that even if you change the draft, and even if HAHA 
home networks don't distribute BGP routes in the core Internet, there is 
no guaranteed globalHAHA benefit in terms of shorter paths?  I think 
there can be many cases where the MR is closer to HA2 (than to HA1) but 
still the path CN-HA1-MR is much shorter than CN-HA2-MR.

GlobalHAHA benefits make strong assumptions not only on the placements 
of the HAs, but also on the placement of the CNs.  It assumes that _if_ 
CN is placed at a certain position then it's better for MN to use the HA 
closer to it.

But placement of CN is really unpredictable.  In some unfortunate case 
CN can be placed such that even if MR uses what it thinks to be its 
closest HA, the whole path CN-HA-MR would be shorter if the MR used the 
HA _farthest_ from it.

>> Now maybe may idea about how to deploy this is indeed stupid, but what
>> can i say, this is what i could infer from what i read in the draft and
>> the comments in the mailing list.... probably i am too short...
> 
> I never understood that you thought it should be deployed that way,
> quite the contrary... And I agreed already to change the draft according
> to your comments... I expect only minor impacts on the text, though...

Would it be possible to insert a section in the draft with an 
illustration of the deployment case where globalHAHA forces MR to use 
the HA closest to it and CN is placed such that MR had better used the 
HA farthest from it?

Alex