<<< Date Index >>>     <<< Thread Index >>>

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





Begin forwarded message:

From: Karl Auerbach <karl@xxxxxxxxxxxx>
Date: January 10, 2006 3:52:31 PM EST
To: David Farber <dave@xxxxxxxxxx>
Subject: Re: [IP] response from Google to yet another twist
Reply-To: Karl Auerbach <karl@xxxxxxxxxxxx>


On Tue, 10 Jan 2006, David Farber wrote:

From: David Presotto via RT <support@xxxxxxxxxxxxxx>

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.

What do you think?   Are we doing something bad?

Well, apart from the notion of doing experiments on unknowning subjects...

Latency is only one of several factors that one wants to know when determining "proximity" on the net when selecting a peer and path to that peer.

For example, if one is proposing to do a large video transfer then latency is not really important. More important is the bandwidth of the path and its ability to handle that bandwidth without packet loss during a TCP flow. So for big video you want a content source that you can reach with a fat clean pipe even if it has somewhat longer latency than a skinny or noisy pipe.

If one is proposing to do a VOIP call, then the desired path not only has latency must also have low variation in that latency (i.e. low jitter.) Thus for voip a low latency+jitter path that is skinny is preferred over a fatter path that has worse latency+jitter.

A while back I created a draft of a protocol to find the numbers that are needed - and yes, there are still gaps in the protocol. But it would, I believe, blow the socks off path and peer selelection using simple latency.

See my blog entry "What Is The Internet Distance From Hither To Yon?" at http://www.cavebear.com/cbblog-archives/000151.html

At the end of that blog entry there's a pointer to the year 2000 note that describes what I called the Fast Path Characterization protocol. (FPCP).
(Shortcut - http://www.cavebear.com/fpcp/fpcp-sept-19-2000.html)

                --karl--







-------------------------------------
You are subscribed as roessler@xxxxxxxxxxxxxxxxxx
To manage your subscription, go to
 http://v2.listbox.com/member/?listname=ip

Archives at: http://www.interesting-people.org/archives/interesting-people/