[e2e] QoS vs Bandwidth Overprovisioning

[Rajesh Talpade]

While it is indeed true that bandwidth in general is available in plenty
at the current moment, there may still be a place for QoS. QoS does not
need to imply the "multiple handshake, back-slapping, etc" protocols
that network elements need to go thru to "reserve resources." It can be
as simple as provisioning network links to support a few traffic classes.
Each traffic class may have different packet loss, delay, jitter
requirements. These can be satisfied by marking traffic to appropriate
traffic classes, edge policing/shaping, provisioning link buffer space,
and using fair queuing to assure minimum bandwidth to each class. 

Clubbing different application traffic together and throwing bandwidth
at it may be a least-common-denominator approach. Link utilization may
have to be kept artificially low such that requirements of all classes
are satisfied, rather than selectively having high utilization for some.
This may motivate ISPs that need to increase utilization across
their network to consider such minimal "QoS" approaches.

Ciao.
Rajesh.




> 	It isn't clear to me that it is more expensive
> to over-provision bandwidth in a backbone than to deploy
> QoS in that backbone.  Some folks here seem to be asserting
> that it is generally cheaper to deploy QoS.
> 
> 	Many ISP folks believe that it is cheaper to over deploy
> capacity, else there would not be so many ISPs that are
> doing so.  Most (maybe all) of the Tier-1 ISPs [1] indicate
> that they over-provision their network and generally engineer
> it such that they don't lose packets inside their own backbone
> (though some packets get lost at the handoff between backbones,
> in my personal experience; pardon the purely anecdotal interjection).
> 
> 	For many big providers, the recurring operations costs 
> (e.g. NOC staff, deployment staff, troubleshooting staff, 
> more generally: hiring and retaining clue) are believed to dwarf 
> the capital costs of building a network -- iff one already has
> dark fibre facilities in place.  This belief might relate to the
> current lack of widely deployed QoS in large backbone networks.
> 
> 	As to the costs of over-provisioning, once one has access
> to dark fibre, it is not particularly more expensive (in capital
> costs) to light it with Gigabit Ethernet than 10baseF or 100baseFX
> (or sundry fibre-optic extenders).  Similarly it isn't really
> that much more expensive to light up a segment with OC-48 than
> OC-12 or OC-3.  My guess is that within 12 months time, lighting
> at OC-192/10 GigE won't be hugely more expensive than OC-48 was
> last year.  It seems more common that big providers have
> access to dark fibre than the smaller providers, so this might
> mean that smaller providers have more incentive to deploy QoS;
> I'm not sure.
> 
> 	As a sidebar, I note that the cost of dark fibre facilities 
> trans-continent has dropped dramatically in North America in the 
> past few years.  Similar cost drops appear to be starting in Europe 
> as well.  Trans-Atlantic capacity has dropped substantially in cost;
> trans-pacific capacity appears poised to have a similar cost
> drop.
> 
> 	So it just isn't clear to me that the marketplace is
> going to drive the facilities-based providers to deploy end-to-end
> QoS anytime particularly soon.  Maybe I'm totally wrong.
> Kindly note well that this is a set of observations about
> what appears to be happening, not a set of assertions about 
> my personal druthers if I had a magic wand.
> 
> Ran
> rja at inet.org
> 
> [1] A market definition of "Tier-1 ISP" might be the set of
> folks, including UUnet, who don't pay money to UUnet to exchange
> traffic.  A more technical definition (slightly different set
> of providers, btw) might be based on having at least dual OC-48 
> capacity cross-continent and a presence in all of the major 10-20 
> metro areas.  For the purpose of this note, the precise definition
> isn't so important as the general concept...
>