Hi Sebastian,

On Friday, July 12, 2013, Sebastian Kiesel wrote:

> Hi Richard,
>
> On Fri, Jul 12, 2013 at 11:41:30AM -0400, Y. Richard Yang wrote:
> > Hi Sebastian,
> >
> > On Fri, Jul 12, 2013 at 10:14 AM, Sebastian Kiesel 
> > <[email protected]<javascript:;>
> >wrote:
> >
> > > Qiushi,
> > > Richard,
> > >
> > > On Fri, Jul 12, 2013 at 09:44:56AM -0400, Y. Richard Yang wrote:
> > > > > Section 1.2.2
> > > > >
> > > > > It's not obvious that the measurement overhead can be reduced by
> > > > > conducting only fine-tuning or fault-tolerant measurement. It's
> better
> > > to
> > > > > list the parts of the overhead which can be removed.
> > > >
> > > > To avoid the details and the complexity that follows, we can remove
> the
> > > > word fine-tuning and say only that the application can use ALTO info
> > > about
> > > > latency and bandwidth, if provided, and hence no need to measure the
> > > > latency or bandwidth itself. How does this sound?
> > >
> > > I think the word fine-tuning shouldn't be removed completely. But maybe
> > > we need some fine-tuning of that paragraph :-)
> > >
> > > The motivation for this paragraph was to say that, in the P2P scenario,
> > > if you really want to know throughput values you will have to measure
> > > them yourself (in the overlay). ALTO can give some predictions who
> could
> > > be good other peers and who not, but there are may reasons why ALTO may
> > > not be accurate (see sec. 8.2.3. of draft-ietf-alto-deployments-06).
> > >
> > > One of the main benefits of ALTO is that if I join a new swarm, ALTO
> may
> > > give me some promising ("better-than-random") peers to start
> > > communicationg with, but then I should do some fine-tuning of my peer
> > > lists based on own measurements.  So the saving of overhead is probably
> > > not saving traffic for probe messages (many measurements are done on
> the
> > > regular application traffic anyway) but it is faster convergence to a
> > > good state.
> > >
> > >
> > Good discussion. Please see below for the revised text:
> >
> >         An application that uses an ALTO Service can benefit from
> >         better knowledge of the network to avoid network
> >         bottlenecks. For example, an overlay application can use
> >         information provided by the ALTO Service to avoid selecting
> >         peers connected via high-delay links (e.g., some
> >         intercontinental links).
>
>
>
> >         Instead of using third-party
> >         databases such as geo-location databases, the application can
> >         use ALTO as a standard to obtain related information.
>
> I don't like this sentence, as it somehow mingles who provides
> information ("third-party" -- we have an own paragraph on who could
> possibly become an ALTO provider) and through which standard
> interface/protocol (ALTO vs. (proprietary) databases).
>
> Maybe just omit it.
>
>
Using ALTO as a unifying interface is a benefit of a standard. But it is
indeed better that we delete the sentence.


>
> >         Using
> >         ALTO to start each node with promising
> >         ("better-than-expected") peers, an adaptive peer-to-peer
> >         overlay may achieve faster, better convergence.
>
> Maybe replace "start" with "initialize".
>
>
OK.


> What is "better-than-expected"?  Please repeat the wording
> "better-than-random peer selection" from the ALTO problem statement
> document.


Sure. It was a typo.

Thanks!

Richard

>
>
> Thanks
> Sebastian
>
> p.s.: thanks to Qiushi for the review!
>
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto

Reply via email to