Hi Ken,
Out here we typically see a ethernet ATM switch and a BDSL tail circuit,
or a fibre link with a 'branch-office grade' switch as the PE. Both have
a 100Mbps ethernet link to the CE, but the ingress rate limit might be
256k-2Mbps (BDSL) or 2-10Mbps or more (fibre). In a customer WAN
environment with 10-30ms delay between sites there usually aren't many
packets outstanding, so TCP's window sizing can't do it's thing, because
the rate-limit discards are effectively occur randomly - longer network
delays can mask that, though. I can't say that I have ever seen an
ingress-rate-limited port at the provider edge flow-control the customer
edge equipment. :)
Real time UDP streams like voice & video actually do a lot better
because they don't retransmit/error correct, and the user usually
selects a more appropriate codec that minimises the dropouts (in other
words the user does manual traffic shaping!)
davidt
On Fri, Dec 23, 2011 at 10:27:07AM -0600, Ken Norton wrote:
> Rate limiting ports is done through flow-control. This is enherent in the
> very nature of ethernet. Has to be otherwise you'd never be able to move
> data through the network which each link is operating at a different speed.
> TCP-IP is quite effective in this manner at handling not only flow-control,
> but packet loss and resends.
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|