On 2008-8-5, at 2:40, ext Iljitsch van Beijnum wrote:
Hm, it looks like the high water marks for TCP and UDP sessions per
subscriber (including non-active subscribers) is around 20 and 40.
http://www.wand.net.nz/~salcock/someisp/flow_counting/result_page.html
The graph you commented on talks about "peak sessions per subscriber
(mean)", and if I understand the text correctly, that means dividing
all active sessions by all active subscribers at the time. It's not
actually a per-user measure.
If you scroll down on that page, the "peak tcp session per active
subscriber (percentiles)" graph shows you how heavy-tailed the
underlying distribution is - 5% of users have more than 100
connections active, and 1% between 500-1000. If a CGN enforced a limit
of 100 connections, an ISP would get calls from 5% of his customers -
probably not what they'd like to see happen.
(These are interesting numbers, by the way - I wish other ISPs could
share similar data!)
Lars
_______________________________________________
Int-area mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/int-area
