[e2e] Skype and congestion collapse.

[David Andersen]

On Mar 6, 2005, at 12:02 PM, Cottrell, Les wrote:

> Great question. Though only a small sample, we also have made  
> measurements to about DSL/ISDN/Cable 10 routers (5 ISPs) in homes in  
> the SF Bay Area for several years now. We have data going back to Jan  
> 1988.  The median losses have dropped from about 3% to about 0.02%,  
> see  
> http://www.slac.stanford.edu/xorg/icfa/icfa-net-paper-jan05/ 
> broadband.jpg

We've also seen this kind of reduction on the RON testbed (1/3  
academic, with a few cable modems and DSL lines).  I haven't peeked at  
the latest numbers, but we saw the average packet loss rate drop from  
0.74% in 2001 to 0.42% in 2003, and my recollection of the more recent  
numbers is that things had improved a bit more.  Our numbers are fairly  
consistent with what Les's broadband graph shows, though there's noise  
in all of the measurements.

(stats taken from an IMC 2003 paper,
  "Best-Path vs. Multi-Path Overlay Routing," by Andersen, Snoeren, and  
Balakrishnan.
  http://nms.lcs.mit.edu/papers/index.php?detail=126 )

Oh - and in. re: the later note about fairness and games, it strikes me  
that these days, home users mostly have themselves to hurt by being  
"unfair," since in the majority of situations, their own link is the  
bottleneck.  That's obviously not always true (overloaded servers,  
cable plants that depend on TCP fairness, if they do, or just  
overloaded ISPs), but for the most part, if you hack your TCP, you're  
just changing the ways your own bandwidth is allocated between your  
flows.  _for home users._

   -Dave

> -----Original Message-----
> From: Jon Crowcroft [mailto:Jon.Crowcroft at cl.cam.ac.uk]
> Sent: Sunday, March 06, 2005 8:13 AM
> To: Cottrell, Les
> Cc: cannara at attglobal.net; End-2-End list; Jon.Crowcroft at cl.cam.ac.uk
> Subject: Re: [e2e] Skype and congestion collapse.
>
> it would be interesting to compare figures for commercial dialup and  
> broadband ISPs wit hacademic - certainly when planetlab is used to  
> study the net in terms of loss/throughput and connectivity, you can  
> see how unrepresentative it is from The Interdoman Connectivity of  
> PlanetLab Nodes. Suman Banerjee, Timothy G. Griffin, Marcelo Pias.  
> Passive and Active Measurement Workshop (PAM) 2004.
> (papers accesible via
> http://www.pam2004.org/
> for interest)
>
>
>
>
> In missive  
> <35C208A168A04B4EB99D1E13F2A4DB01A0D71B at exch- 
> mail1.win.slac.stanford.edu>, "Cottrell, Les" typed:
>
>>> Our measurements are using ping mainly between academic and research  
>>> sites worldwide. The high losses in the US was several years ago.  
>>> Typical losses between developed regions of the world (Canada, US,  
>>> Europe, Japan, Korea, Singapore, Australia/NZ) are << 1%. Things  
>>> have improved a lot. See for example Figs 3 and 4 of  
>>> http://www.slac.stanford.edu/xorg/icfa/icfa-net-paper-jan05/.  There  
>>> are high losses still in parts of the Internet but they are mainly  
>>> to developing countries (see for example Figs 9 and 10 of the above  
>>> report).
>>>
>>> -----Original Message-----
>>> From: end2end-interest-bounces at postel.org  
>>> [mailto:end2end-interest-bounces at postel.org] On Behalf Of Cannara
>>> Sent: Saturday, March 05, 2005 9:43 PM
>>> Cc: End-2-End list
>>> Subject: Re: [e2e] Skype and congestion collapse.
>>>
>>> On where congestion is occurring today in the Inet, it would be  
>>> useful to ask Les Cottrell at SLAC how things have been going with  
>>> their continued "ping-around-the-world" testing.  It had been true  
>>> that busy peering points in the East & West of the US commonly lost  
>>> 30%.
>>>
>>> Alex
>>>
>>> RJ Atkinson wrote:
>>>>
>>>> On Mar 4, 2005, at 17:47, Clark Gaylord wrote:
>>>>> This is why we really do need some notion of QoS other than The Fat
>>>>> Pipe.  It doesn't have to be as elaborate as RSVP-disciplined CAC,
>>>>> but you need to be able to prioritize traffic that matters and  
>>>>> limit
>>>>> the amount of traffic that gets prioritized.  It doesn't have to be
>>>>> more complex than that, but it has to do at least that.  [Ergo ...
>>>>> left as an exercise to the reader.]
>>>>
>>>> I don't know that the "network" needs to have a more sophisticated
>>>> notion of QoS than best effort.  It can sometimes be useful for the
>>>> network device connected directly to a congested link (e.g. access
>>>> link between a site and its upstream provider) to have some
>>>> internal-to-the-box QoS configuration.
>>>>
>>>> It is not uncommon these days for the access router at the customer
>>>> premise to have some ACL ruleset that prefers some traffic over  
>>>> other
>>>> traffic or rate-limits certain kinds of traffic -- and equivalent
>>>> configuration of the aggregation router on the ISP side of the same
>>>> link is also not uncommon these days.
>>>>
>>>> That said, most congestion today occurs either on an access link  
>>>> such
>>>> as that or on some sort of wireless link (e.g. SATCOM to SW Asia).
>>>> ISP core backbones tend to be over-provisioned.  Most campus
>>>> (wired/fibred) networks are similarly over-provisioned.
>>>>
>>>> Ran
>
>  cheers
>
>    jon
>
>
>