[e2e] Queue size of routers
David P. Reed
dpreed at reed.com
Tue Jan 21 06:51:45 PST 2003
At 09:09 AM 1/21/2003 -0500, RJ Atkinson wrote:
> A more direct and proximal cause for the low probability of backbone
>link congestion is that most large ISPs deliberately over-engineer their
>backbones to have visibly more capacity than the max offered load. No amount
>of flow smoothing could counteract congestion caused by an absence of
>over-provisioning of backbone links.
Just a thought about terminology that may confuse in this paragraph.
"max offered load" is correct, but some may read it as "max possible load"
- I don't think most backbones are engineered to exceed any possible load
that can be applied. Instead real experience is used to determine what a
peak load looks like.
"over-provisioning" and "over-engineering" should really be deprecated as
terms because they sound negative, even unprofessional or inefficient. In
fact, the provisioning of extra capacity is one of the least expensive ways
in most cases to achieve good quality of service (not QoS, but the ordinary
lay understanding of that phrase). We should reserve the terms
over-engineering and over-provisioning to those cases where no user benefit
accrues. It is not "over-engineering" to deploy a 10 Mb/s Ethernet hub
where the load on the network averages out to 50 kb/s, for example.
More information about the end2end-interest