interesting-people message

[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [Elist Home]

Subject: [IP] more on response from Google to yet another twist

  • From: David Farber <>
  • To:
  • Date: Tue, 10 Jan 2006 15:01:06 -0500

Begin forwarded message:

From: Bob Frankston <>
Date: January 10, 2006 2:48:14 PM EST
Cc: &quot;'David Presotto via RT'&quot; <>
Subject: RE: [IP] response from Google to yet another twist

Such information is very very valuable but it could be done on a far larger scale openly.

The SETI project worked because it was a trusted organization that wasn't trying to monetize something.

If people could trust a simple application that cooperated to create a dynamic map of Internet performance measurements -- speed, latency etc against routes, time of day, carriers and other factors, you'd see many people want to cooperate in return for having the information available.

Right now I can run speed tests using something like http:// but it shows only the speed. That alone is interesting &#x2013; I&#x2019;ve compared RCN, Comcast and Verizon. What most concerns me is that, on FIOS, I can get a 15mbps speed from and 1.6mbps to New York but the speeds to the West Coast drop off sharply. I notice similar performance on the other carriers (though RCN is about 10mbps and Comcast about 6 or 7mbps down). But I don&#x2019;t know how much is due to speakeasy and what the numbers are to rest of the world. I don&#x2019;t know the latency and other detailed information.

While such measurements do exist a very widely distributed system of responders at the edge on consumer systems could provide a very useful and dynamic map of the real net and let the users identify anomalies. The measurements won&#x2019;t necessarily tell us what is possible &#x2013; just what shows through current policies and implementations. It could also help detect games carriers may be playing.

I don&#x2019;t know how well it would work in practice but the kind of measurements being described here are just part of such an effort. Being more transparent and scaling it up could be interesting. One caution &#x2013; do use some techniques to minimize the measurements themselves being gamed. The software can also try to identify the particular router/NAT versions along the way.

-----Original Message-----
From: David Farber []
Sent: Tuesday, January 10, 2006 13:34
Subject: [IP] response from Google to yet another twist

Sounds reasonable. Any other comments??


Subject: Re: [PL #12596] external publicity for google_highground

Date: Mon, 9 Jan 2006 23:12:07 -0500

From: David Presotto via RT <>



Email Recipients (see

        Owner: Nobody


        Ticket Ccs:, google_highground@slices.planet-


Hi Dave,

Long time no see.

The hits are a result of an experiment set up by a summer

intern, Michal Szymaniak.   We're trying to latency map the

internet using a 6 dimensional model he set up for his PhD


The idea is to use some number of reference points (planet

lab nodes, google datacenters, etc) and measure latencies from

google customers to them.   We then use those latencies to try

to place the nodes in the 6D space.   Once we've mapped

enough /24s, we can guess the latency from any IP to any other

IP with a 10 to 15% error.  More importantly for us, we can cluster

the IPs around our datacenters so we can always point users

at the closest datacenter.   (For that we have to also map all

users to the DNS resolvers they use since the only steering

mechanism we have is via DNS, but that's another experiment...)

The way we're getting users to touch the different reference

points is using the header in firefox:

      Link: url options

If you use the option 'prefetch', firefox does an access of

the link referenced in background.

In our case the thing accessed is a page that just says something

like 'thank you for participating in the google latency experiment'.

We record what IP accessed and the latency time (taken from

SYN-ACK to ACK time).   Since we already know the IP our

customers use to access us, that doesn't give us much more

identity info.   It does let us locate them in space a bit better than

we currently know from services like Quova though.

The results will be part of Michal's thesis.   Right now we're not

using it

ourselves, though we may put up an interface for anyone to use

that lets them type in two IPs and get an estimated latency.   I'm

considering turning off the planetlab part if he now has enough

data.   We're happy to use just Google datacenters as reference

points for ourselves.

What do you think?   Are we doing something bad?

On Jan 9, 2006, at 6:43 PM, Steve Muir wrote:

> Dear Google folks,


> a Google user who's noticed the connections to PlanetLab nodes

> being made

> has posted the message below to Dave Farber's IP list, which has a

> pretty

> broad readership.  please can one of you respond to Dave Farber

> with an

> explanation of what's going on.  i know you've told us before but

> i'll let

> you give them the official version.


> thanks,


> steve




> On Mon, 9 Jan 2006, Steve Muir via RT wrote:



>> Date: Mon, 9 Jan 2006 20:42:27 -0500

>> From: David Farber <>

>> To:

>> Subject: [IP] more on ****yet*****  Another new Google twist..




>> Begin forwarded message:


>> From: Rodney Joffe <>

>> Date: January 9, 2006 6:11:13 PM EST

>> To: Lauren Weinstein <>, Dave Farber

>> <>

>> Subject: Re: [IP] more on Another new Google twist..


>> Hi Lauren,


>> On Jan 9, 2006, at 1:36 PM, David Farber wrote:




>>> Begin forwarded message:


>>> From: Lauren Weinstein <>

>>> Date: January 9, 2006 2:41:01 PM EST

>>> To:

>>> Cc:

>>> Subject: Re: [IP] Another new Google twist..


>>> This is not entirely a straightforward situation.  First, such

>>> history displays are almost certainly based on cookies, so persons

>>> who do not allow Google cookies are unlikely to see such output.

>>> (Note however that this is a separate issue from Google's internal

>>> logs of user search activity presumably tied to IP addresses.)


>> But wait, there's more. I have also been noticing seemingly random

>> but frequent attempts to trigger firefox connections to various

>> planetlabs machines ( as a result of

>> Google searches. I think it is admirable that Google is supporting

>> the research world but were it not for my &quot;littleSnitch&quot; application,

>> I would have had no idea. Nor do I know what Google is triggering, or

>> what data is being forwarded to the planetlabs network, or why - I

>> haven't bothered to stop it thus far. Have you noticed this?

>> &quot;Googling&quot; for this brings up general hints showing Google's

>> involvement, but I can't find any official note in Google's help

>> pages or FAQ.


>> Machines include:

>> Server: (

>> Server: (

>> Server: (

>> Server: (

>> Server: (



>> ----------------------------------------------

>> Rodney Joffe

>> CenterGate Research Group, LLC.


>> &quot;Technology so advanced, even we don't understand it!&quot;(R)


>> _______________________________________________

>> PlanetLab Support Mail Reflector






PlanetLab Support Mail Reflector


You are subscribed as

To manage your subscription, go to

Archives at: people/

You are subscribed as
To manage your subscription, go to

Archives at:

[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [Elist Home]

Search: Match: Sort by:
Words: | Help


Powered by eList eXpress LLC