Re: [p2pi] Information in an ALTO protocol

On Fri, Sep 12, 2008 at 3:40 PM, Stanislav Shalunov <shalunov@shlang.com>wrote:

> Let us put the privacy considertaions aside for a minute.
>
> I'm actually not even sure how one would use uplink sizes if they were
> available, particularly only for some peers.
>
> The underlying assumption of many messages seems to be that P2P clients
> will prefer to connect to higher-uplink peers.  While this would help you if
> you're the only one doing it, it would hurt everyone if done universally.

[Ye (Yale)]
We have seen many studies on video streaming applications that prefer
higher-uplink-bandwidth peers.  I certainly agree with you that the system
may somehow get 'blocked' if all clients naively select the highest-uplink
peer to be parents.  How to efficiently use all upload capacity, especially
those higher-uplink peers, is still an interesting problem.

>
>
> The only use I can think of right now that doesn't have obvious adverse
> consequences is to prefer peers with uplinks of about the same size as
> yours.  While this shouldn't reduce average download rate, I'm also not sure
> what purpose it would serve.  It would make the response to increases in
> uplink capacity more linear, but of course it's sublinear by design.
>
[Ye (Yale)]
Analysis based on a fluid model of P2P file sharing system indicates that
bandwidth matching can help improve average file download rate.  The
bandwidth matching can be roughly described like: a peer with download
capacity R downloads from the peer whose upload capacity R (=downloader
download rate).  Thus, for file sharing applications, like Pando,
BitTorrent, etc., the uplink/downlink bandwidth information can be useful in
system optimization.
In the recent P4P field test, we estimated these numbers before using them
to compute peering selection weights.  Estimation based on measurement is
dynamic and less accurate, while direct numbers from ALTO servers may be
generally static and more accurate.

>
> Can someone fill me in on how a P2P app would use provisioned uplink
> numbers?
>

> Thanks, -- Stas
>
> -----Original Message-----
> From: "Laird Popkin" <laird@pando.com>
> To: "Salman Abdul Baset" <sa2086@columbia.edu>
> Cc: p2pi@ietf.org; "Richard Woundy" <Richard_Woundy@cable.comcast.com>
> Sent: 9/11/08 1:28 PM
> Subject: Re: [p2pi] Information in an ALTO protocol
>
> (1) If an ISP configures their ALTO server to tell applications that the
> users' uplink is less than it really is, I think that could have two
> effects. For users inside the ISP, it would tend to suppress internal
> traffic and cause peers to download from external peers, which is probably
> bad for the ISP. And it might tend to cause peers from outside the ISP to
> shift traffic elsewhere to some degree, which would reduce uplink
> utilization, which would be an incentive to 'aim low'. How these two factors
> would balance out would depend on the ISP's infrastructure.
>
> (2) I don't believe that the p2p network would have to decide that the
> ISP's ALTO server was 'lying'. P2P networks all currently measure observed
> throughput between peers and use that to make decisions. If they add ALTO
> provisioning information, I would expect that they would treat observed
> throughput as being more true than the ALTO guidance. So without having to
> make a decision about 'ALTO lying' the p2p network would tend to take
> advantage of good ALTO guidance (because ALTO guidance and p2p performance
> optimization reinforce each other) and would tend to cancel out bad ALTO
> guidance (because the p2p performance optimization would end up overruling
> the ALTO guidance).
>
>  - Laird Popkin, CTO, Pando Networks
>  mobile: 646/465-0570
>
> ----- Original Message -----
> From: "Salman Abdul Baset" <sa2086@columbia.edu>
> To: "Laird Popkin" <laird@pando.com>
> Cc: "Richard Woundy" <Richard_Woundy@cable.comcast.com>, p2pi@ietf.org
> Sent: Thursday, September 11, 2008 3:46:10 PM (GMT-0500) America/New_York
> Subject: Re: [p2pi] Information in an ALTO protocol
>
> This raises at least two issues:
>
> > Playing devil's advocate, while ALTO doesn't provide a mechanism for ISP
> A
> > to tell non-customers anything, but it is possible for the p2p network
> > as a whole to make decisions based on information provided by ALTO. For
> > example, if ISP A says that its users have bad uplink, and ISP B says
> > that its customers have great uplink, then the rest of the internet will
> > tend to download more from ISP B than A.
>
> (1) Do ISPs have an incentive to lie to their customers that their uplink
> is bad?
>
> This will be counter-balanced
> > in that if, in practice, ISP A has fine uplink, the p2p network will
> > figure that out and use those peers anyway, ignoring the misleading ALTO
> > guidance, because observed behavior has (IMO) more 'weight' than ALTO
> > guidance.
>
> (2) Do P2P applications must determine if an ISP ALTO server is lying?
>
> Note, that misconfiguration can also be a cause of ALTO server providing
> misleading information.
>
> -s
>
>
> >
> > ----- Original Message -----
> > From: "Richard Woundy" <Richard_Woundy@cable.comcast.com>
> > To: "Reinaldo Penno" <rpenno@juniper.net>, "Laird Popkin" <
> laird@pando.com>
> > Cc: p2pi@ietf.org
> > Sent: Thursday, September 11, 2008 2:52:06 PM (GMT-0500) America/New_York
> > Subject: RE: [p2pi] Information in an ALTO protocol
> >
> >> I was wondering if he was assuming some kind inter-ALTO communication.
> >
> > I guess I was. And perha
>  _______________________________________________
> p2pi mailing list
> p2pi@ietf.org
> https://www.ietf.org/mailman/listinfo/p2pi
>