Right, but that is for DHAAD. I thought the proposal here was for the MN to
send the IKE_SA_INIT to the anycast address, for the anycast address to
forward it to an HA, then for that HA to respond.
jak
----- Original Message -----
From: "Vijay Devarapalli" <vijay.devarapalli@xxxxxxxxxxxxx>
To: "James Kempf" <kempf@xxxxxxxxxxxxxxxxxx>
Cc: "Kilian Weniger" <Kilian.Weniger@xxxxxxxxxxxxxxxx>; <mip6@xxxxxxxx>
Sent: Tuesday, July 25, 2006 11:51 AM
Subject: Re: [Mip6] New charter item: HA assignment
James Kempf wrote:
Combining anycast with DNS is certainly an option. We could put something
about that into the split draft. Anybody else have an opinion on that?
the MN using DNS to obtain the anycast address for the home
agents is briefly mentioned in RFC 3775. see appendix A.5.
but we should elaborate more on this in the split scenario
solution draft.
Vijay
But I think there is a general problem with using anycast for server load
balancing, or in any case where the measure for which server gets the
traffic isn't some routing-based metric. This isn't something the MIP6 WG
can take up, though.
jak
----- Original Message ----- From: "Kilian Weniger"
<Kilian.Weniger@xxxxxxxxxxxxxxxx>
To: "Vijay Devarapalli" <vijay.devarapalli@xxxxxxxxxxxxx>; "James Kempf"
<kempf@xxxxxxxxxxxxxxxxxx>
Cc: <mip6@xxxxxxxx>
Sent: Tuesday, July 25, 2006 1:48 AM
Subject: RE: [Mip6] New charter item: HA assignment
Right, I assumed a more or less stable anycast group. BTW, I'm not
against the DNS solution per se. It may make sense to combine it with
other methods, e.g., if anycast is used, different anycast HA addresses
could be obtained from DNS for different mobility services
("_mip6._ipv6.example.com" for MIPv6, "_nemo._ipv6.example.com" for
NEMO, "_dsmip6._ipv6.example.com" for DSMIPv6 etc.).
-Kilian
-----Original Message-----
From: Vijay Devarapalli [mailto:vijay.devarapalli@xxxxxxxxxxxxx]
Sent: Montag, 24. Juli 2006 21:17
To: James Kempf
Cc: Kilian Weniger; mip6@xxxxxxxx
Subject: Re: [Mip6] New charter item: HA assignment
James Kempf wrote:
> I thought the whole point of the anycast proposal and Kilian and
> Jean-Marie's argumet against DNS was load balancing and that DNS is
> insufficient because it does not provide fast enough
dynamic update. If
> that is the case, then I would think that the ability to
quickly add and
> remove addresses from the anycast group would be crucial to
removing HAs
> that are loading and adding new ones that aren't.
load balancing does not necessarily mean removing/adding a HA
from/to the list of HAs. I think it is more about directing the
users to the right HA. you can have a pretty stable anycast
group and still direct the MN to a specific HA.
Vijay
>
>
> jak
>
> ----- Original Message ----- From: "Vijay Devarapalli"
> <vijay.devarapalli@xxxxxxxxxxxxx>
> To: "James Kempf" <kempf@xxxxxxxxxxxxxxxxxx>
> Cc: "Kilian Weniger" <Kilian.Weniger@xxxxxxxxxxxxxxxx>;
<mip6@xxxxxxxx>
> Sent: Monday, July 24, 2006 12:08 PM
> Subject: Re: [Mip6] New charter item: HA assignment
>
>
>> hi Jim,
>>
>> I think manual configuration is sufficient in this case. it is
>> not too often that home agents get added to or deleted from the
>> the anycast group.
>>
>> Vijay
>>
>> James Kempf wrote:
>>> The other problem with this solution is that there is
still no secure
>>> way for a collection of unicast server addresses to get
assigned to
>>> an anycast address, except for static configuration. So
how would the
>>> operator be able to add HAs to the anycast address?
>>>
>>> At least, with the DNS solution, the update of the DNS can be
>>> automated using dynamic DNS.
>>>
>>> jak
>>>
>>> ----- Original Message ----- From: "Vijay Devarapalli"
>>> <vijay.devarapalli@xxxxxxxxxxxxx>
>>> To: "Kilian Weniger" <Kilian.Weniger@xxxxxxxxxxxxxxxx>
>>> Cc: "James Kempf" <kempf@xxxxxxxxxxxxxxxxxx>; <mip6@xxxxxxxx>
>>> Sent: Monday, July 24, 2006 10:26 AM
>>> Subject: Re: [Mip6] New charter item: HA assignment
>>>
>>>
>>>> Kilian Weniger wrote:
>>>>
>>>>> However, the problem depends on how the initial HA is selected.
>>>>> Anycasting can be used to prevent that new MNs select
the same initial
>>>>> HA. For instance, the IKE_SA_INIT can be anycasted.
>>>>
>>>> is there another instance of using anycast for load balancing.
>>>> just curious.
>>>>
>>>>> But if this is done,
>>>>> you can easily get rid of the unnecessary SA
establishment and the
>>>>> switch message by requiring the receiver HA to forward
the IKE_SA_INIT
>>>>> to a MN-specific and low-loaded HA (see
>>>>
>>>> how do you do this? you would still need a HA-to-HA protocol to
>>>> forward the IKE_SA_INIT message to another HA. when you send a
>>>> message to an anycast address routing decides which HA receives
>>>> the message.
>>>>
>>>> I have seen anycast used to direct a user to the
"nearest" server.
>>>> "nearest" means nearest by routing hops. when all the HAs on the
>>>> same link, I wonder if anycast gets any load balancing at all.
>>>>
>>>> Vijay
>>>>
>>>>> draft-dupont-ikev2-haassign-01.txt for details).
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>>
>>>>> -Kilian
>>>>>
>>>>>> jak
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> Mip6 mailing list
>>>>>> Mip6@xxxxxxxx
>>>>>> https://www1.ietf.org/mailman/listinfo/mip6
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Mip6 mailing list
>>>>> Mip6@xxxxxxxx
>>>>> https://www1.ietf.org/mailman/listinfo/mip6
>>>>
>>>>
>>>
>>>
>>
>>
>
>
