Interesting bandwidth usage information. Here is a related draft on bandwidth provisioning and management from Comcast ( http://www.ietf.org/id/draft-livingood-woundy-congestion-mgmt-01.txt): when a user pumps traffic at 70-80% of provisioned thresholds for 15 minutes, his/her packets will be added to a low-priority queue that may be dropped in event of congestion.
I guess in the context of P2P, the tracker manager needs to swap seeds every 15 minutes. :-) - Ping On Wed, Feb 10, 2010 at 1:14 PM, Vijay K. Gurbani <[email protected]>wrote: > Hi: During the Hiroshima IETF, there was some discussion on > how to use the provisioned bandwidth for ALTO [1]. > > I suspect that one reason why there hasn't been a more > in-depth list discussion on this topic after Hiroshima may be > the lack of public availability of specific characteristics > of residential networks. Some aspects we touched upon in > the Hiroshima discussion included the frequency by which > the IP address changes to make it long-term reliable host > identifier, etc. > > At about the same time we were meeting in Hiroshima, the > ACM IMC conference was taking place in Chicago. One of the > papers there is entitled "On dominant characteristics of > residential broadband Internet traffic" by Maier et al. [2]. > The data is based on 20,000 residential DSL lines belonging to > a major European ISP (we probably should stay away from > drawing a general conclusion for other access methods like > cable and fiber.) Regardless, the paper makes for some > interesting reading and provides some insight into DSL > residential broadband. > > The findings in here are interesting and we can use some > of these for our work in ALTO. To summarize some of the > work: > > 1) Less than a quarter of active lines exceed 50% of their > bandwidth for even one second over a 5m period. During the > day, 50-60% of the active lines achieve at least a 10% band- > width utilization. > > 2) Session durations are surprisingly short -- a median > duration of only 20-30m. > > 3) IP addresses are re-assigned frequently, with up to 4% > of addresses assigned more than 10 times a day, and 50% > are reassigned at least twice in 24h (in other words, > the use of IP addresses as host identifiers can be > misleading over fairly short time scales.) > > 4) HTTP dominates (> 50% of traffic) in terms of bytes due > to streaming of Youtube videos, etc. > > 5) Examining TCP round-trip times, for many TCP connections the > access bandwidth-delay product exceeded the advertised windows, > making it impossible for the connection to saturate the > access link. > > [1] Item 3 in http://trac.tools.ietf.org/wg/alto/trac/wiki/Ietf76 > [2] G. Maier, A. Feldmann, V. Paxon, and M. Allman, "On dominant > characteristics of residential broadband Internet traffic," > Proceedings of the 9th ACM SIGCOMM conference on Internet > measurement conference (IMC), pp. 90-102, November 2009. > > Thanks, > > - vijay > -- > Vijay K. Gurbani, Bell Laboratories, Alcatel-Lucent > 1960 Lucent Lane, Rm. 9C-533, Naperville, Illinois 60566 (USA) > Email: v...@{alcatel-lucent.com,bell-labs.com,acm.org} > Web: http://ect.bell-labs.com/who/vkg/ > _______________________________________________ > alto mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/alto >
_______________________________________________ alto mailing list [email protected] https://www.ietf.org/mailman/listinfo/alto
