Le 10 août 2010 à 21:06, Fred Baker a écrit :

> I would find that surprising. There are ample cases where the originator of a 
> high data rate flow (sensor data from a radio telescope to a number cruncher, 
> to pick one example) might want to use the flow label to send data from one 
> session down multiple paths. Multi-path TCP would be another example.

> I suppose you could argue that consistently alternating among N flow labels 
> can be thought of as N separate flows, but if it's TCP there are other 
> ramifications.

Packets that are expected to go down multiple paths should be identified as 
belonging to different flows as far as FLs are concerned.
In the case of multi-path TCP, where "one or both peers are expected to have 
multiple addresses", this is already the case: different paths are taken by 
packets of different MPTCP "subflows", i.e. packets having different 5-tuples.


> 
> In any event, the requirement of the load balancer is not that the originator 
> gives things a tag; we can develop such tags from a hash of the source and 
> destination address quite well, thanks.

Originators of TCP and UDP connections should clearly hash 5-tuples when they 
exist (not 2-tuples in this case as you seem to assume).

> The important thing is that things are tagged in a manner that helps the load 
> balancer.

Agreed.

> There are two major uses of load balancers.

> In data centers, we balance sessions to/from sets of servers, and the 
> question when a new session comes up is which server is most likely to be 
> able to effectively serve it (has the necessary resources including access to 
> the right disks, CPU cycles, etc). In bandwidth allocation uses (spreading 
> load across multiple otherwise-equivalent paths), the question is how to use 
> each of the paths most effectively - we want some traffic on each of them and 
> we want none of the paths to be overloaded or introducing more delay or loss 
> than others. Everyone's instinctive knee-jerk response to the question seems 
> to be "hash something and depend on the law of large numbers to serve the 
> need", and the observation of numerous operators is that the law of large 
> numbers is a good start but is not an adequate solution.

"Hash something" (and possibly a 2-tuple even when a 5-tuple is available) is 
clearly not as good as "hash 5 tuples if available", which I propose to become 
a SHOULD.

Taking a "good start" as starting point, and trying to improve from there, is 
what I try to propose.  


> You really want the load balancer to be able to be smarter than that.
> 
> Which is why people that build load balancers frequently do so using NAT 
> technology or some other way of tagging individual sessions.

Are you arguing in favor of NATs in IPv6, adding load balancing as a reason to 
use them :-(?

As a convinced promoter of IPv6, and of its restored e2e transparency, I find 
that IPv6 load balancing solutions that don't need NATs are worth specifying.

> And why they are looking for mutable flow labels in this thread.

I don't see why 0-FL mutability wouldn't be sufficient for them (provided new 
values must statistically different for different flows, including subflows, 
micro-flows, whatever can be identified).

But more data on what would be their problem are most welcome. 

Regards,
RD

 
> 
> On Aug 10, 2010, at 9:09 AM, Michael Richardson wrote:
> 
>> 
>>>>>>> "Rémi" == Rémi Després <[email protected]> writes:
>>   Rémi> RFC 3697 isn't concerned with ASes, and doesn't need to be.
>>   Rémi> The proposal is only that, where load balancing is performed, 
>>   Rémi> 0 FLs MAY be replaced by meaningful values for this purpose.   
>>   Rémi> A FL, once set to a non 0 value, never needs to be reset.
>> 
>> okay, so what you are saying is that load balancing uses of the FL are
>> only upset when they see zero.  So for instance, if layer-4s (i.e. end
>> points) were mandated that they must now always set a FL consistently on
>> a flow, and set it to a non-zero value, that this would satisfy the
>> requirements of load balancers.
>> 
>> In this context, permission would be given to set zero FL to a non-zero
>> value. 
>> 
>> -- 
>> ]       He who is tired of Weird Al is tired of life!           |  firewalls 
>>  [
>> ]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net 
>> architect[
>> ] [email protected] http://www.sandelman.ottawa.on.ca/ |device 
>> driver[
>>  Kyoto Plus: watch the video <http://www.youtube.com/watch?v=kzx1ycLXQSE>
>>                     then sign the petition. 
>> --------------------------------------------------------------------
>> IETF IPv6 working group mailing list
>> [email protected]
>> Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
>> --------------------------------------------------------------------
> 
> http://www.ipinc.net/IPv4.GIF
> 


--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------

Reply via email to