nanog mailing list archives

Re: Observations of an Internet Middleman (Level3)


From: Scott Helms <khelms () zcorum com>
Date: Thu, 15 May 2014 14:06:24 -0400

Chris,

You're not reading what I said, nor did I make a statement anything like
 one of the silly things you referenced (640k ram etc).  Prioritization
isn't that complex and today we handle the maximum amount of complexity
already since everything is the same priority right now.

You're trying to make the statement that giving multiple content providers
priority somehow makes connectivity unworkable for consumers as if we don't
have this problem already.  Consumers can easily starve themselves of
bandwidth with video or any other content and almost no connections in the
US have any sort of intelligent fair usage buffering provided by the
service provider.  This is true for both cable, telco, and other operators.


Scott Helms
Vice President of Technology
ZCorum
(678) 507-5000
--------------------------------
http://twitter.com/kscotthelms
--------------------------------


On Thu, May 15, 2014 at 2:01 PM, Christopher Morrow <morrowc.lists () gmail com
wrote:

On Thu, May 15, 2014 at 1:48 PM, Scott Helms <khelms () zcorum com> wrote:
Its not really that complex, if you think about it having 10000s of
'movieco' with the same priority is the status quo.  At the end of the
day
the QoS mechanics in DOCSIS are pretty straightforward and rely on
service
flows, while service flows can have equal priority I doubt most operators
will sell more than a few (perhaps just one) top priority in a given a
category.


yes, there will only ever be 5 computers. or you couldn't possibly
need more than 640kb of ram..... or more than 4billion 'ip addresses'.

I don't think you have to get to more than 10 or 20 of the stated
examples before things get dicey ... Once a set of customers
experience (and can measure) the effect, they'll back their complaints
up to 'moviecompany' and some set of contract penalties will kick in,
I suspect.

Sure, if there is only one it's not a problem, but there are already
not just one...


Scott Helms
Vice President of Technology
ZCorum
(678) 507-5000
--------------------------------
http://twitter.com/kscotthelms
--------------------------------


On Thu, May 15, 2014 at 1:22 PM, Christopher Morrow
<morrowc.lists () gmail com> wrote:

On Thu, May 15, 2014 at 1:06 PM, Ryan Brooks <ryan () hack net> wrote:
On 5/15/14, 11:58 AM, Joe Greco wrote:

2) Netflix purchases 5Mbps "fast lane"


I appreciate Joe's use of quotation marks here.    A lot of the dialog
has
included this 'fast lane' terminology, yet all of us know there's no
'fast
lane' being constructed, rather just varying degrees of _slow_ applied
to
existing traffic.


please correct me if I'm wrong, but 'fast lane' really is (in this
example):
  'cableco' port from 'moviecompany' has 'qos' marking configuration
to set all 'moviecompany' traffic (from this port!) to some priority
level.

  customer-port to 'cableco' has 'qos' handling/queuing that will
ensure '5mbps' of 'moviecompany' is always going to get down the link
to the customer, regardless of the other traffic the customer is
requesting.

right? (presume that in the rest of the 'cableco' network is
protecting 'moviecompany' traffic as well, of course)

So, when there are 1 'moviecompany' things to prioritize and deliver
that's cool... but what about when there are 10? 100? 1000? doesn't
the queuing get complicated? what if the 'cableco' customer with
10mbps link has 3 people in the location all streaming from 3
different 'moviecompany' organizations which have paid for 'fastlane'
services?

3 x 5 == 15 ... not 10. How will 'cableco' manage this when their
100gbps inter-metro links are seeing +100gbps if 'fastlane' traffic
and 'fastlane' traffic can't make it to the local metro from the
remote one?

This all seems much, much more complicated and expensive than just
building out networking, which they will have to do in the end anyway,
right? Only with 'fastlanes' there's extra capacity management and
configuration and testing and ... all on top of: "Gosh, does the new
umnptyfart card from routerco actually work in old routerco routers?"

This looks, to me, like nuttiness... like mutually assured destruction
that the cableco folk are driving both parties into intentionally.

-chris

BTW: I didn't use a particular 'cable company' name for 'cableco', nor
did I use a particular streaming media company for 'moviecompany'...
Also, 'cableco' is short-hand for
'lastmile-consumer-provider-network'. Less typing was better, for me,
I thought.





Current thread: