[rrg] Fundamental objections to a host-based scalable routing solution

[Luigi Iannone]

Christian,

my only point was that using host probing does not allow you to see 
which link is down, since your traffic can go through different LOCs 
depending on the direction.

I agree in general with what you said, except that I do not think that 
probing in the network is harder than probing in the hosts, you just 
have  a different viewpoint.

Luigi

Christian Vogt wrote:
> Luigi Iannone wrote:
> 
>> symmetric traffic : is the same remark as Dino, i.e., unidirectional
>> traffic.
>>
>> symmetric routing: your data packet and your ack do not pass necessarily
>> through the same LOCs (which limits the capacity of discover the broken
>> link). Am I wrong?
> 
> Luigi and Dino:
> 
> The problem of asymmetric routing exists independently of whether you
> probe explicitly, or implicitly using payload traffic.  Also, probing in
> hosts (rather than in the network) does not make the problem harder.  In
> fact, the problem gets harder if you probe in the network, because only
> then do you potentially have to synchronize state over multiple boxes.
> 
> Regarding unidirectional traffic:  Yes, this requires explicit probing
> into the silent direction.  And for host-based solutions, explicit
> probing cannot easily be aggregated.  Having said this, I wouldn't take
> it for granted that aggregation of probing traffic in network-based
> solutions will in all situation yield a noticeable reduction in probing
> traffic either.  The number of sessions between a given pair of networks
> may be very small.
> 
> - Christian
> 
>