Re: [alto] Question regarding voluntary nature of ALTO clients

Generally, the answer to the question of how a protocol is policed (or
enforced) is that it isn't (policed or enforced).  There is no "protocol
police" that will impose punitive measures on non-behaving clients.

Also generally speaking, well-behaving clients will have an automatic
incentive to continue to behave well since doing so benefits them in some
manner (shorter latencies, shorter download time, etc.).  Similarly, for
the server to disseminate bogus information (as you point in your email)
implies that the entire system will be unused since clients will not prefer
to use guidance that provides sub-optimal performance.  So from a
cost-effort point of view, the system reaches an equilibrium, whether or
not this is a unique Nash equilibrium or there are multiple Nash equilibria
is subject for an academic paper.  Clearly, if a client (or server) pursues
a dominant strategy and is greedy at all times, the value of the overall
system decreases.

Cheers,

- vijay

On Tue, Jul 7, 2020 at 1:44 PM Paulo Edgar Mendes Caldas <
a79089@alunos.uminho.pt> wrote:

> Good morning,
>
> I am currently doing investigative work regarding the ALTO protocol for my
> master's degree. I've been getting acquainted with the ALTO protocol and
> the problems it tries to solve via reading of the RFCs and drafts that have
> been published, but I've struggled to find a question to the following -
> does the ALTO system as a whole have a plan on how to guarantee that the
> clients act in good faith with the information they receive?
>
>  In particular, consider that a P2P application wishes to have ALTO
> guidance to pick between a number of potential number of peers, and thus
> queries for a multi-cost map for the pairs (source_pid, destination_pid)
> whose throughput and one-way-delay values are within a certain criteria. Is
> there anything preventing said client to then not prefer the pair whose
> routing cost is lowest? It would make sense that the ALTO server would
> prefer that the client would "be kind" and cooperate with the ISP and use
> the information in a way that is mutually beneficial, but what if the
> client only cares about client performance?
>
> I can only imagine one of the following solutions - there being some
> incentive mechanism, restricting the information to trustworthy partners,
> or deliberately manipulating non-routing-cost metric values to steer
> clients into a certain choice. The latter seems problematic as it breaks
> the transparency in layer cooperation and might damage the entire reason
> the system was created to begin with.
>
> Is there any work done in regards to ensuring client cooperative behavior?
>
> Kind regards,
>
> Paulo Caldas
> _______________________________________________
> alto mailing list
> alto@ietf.org
> https://www.ietf.org/mailman/listinfo/alto
>