RE: remember the "diameter of the internet"?

Mailman List Mirror (Read Only)
1

Regarding the diameter of the Internet - I'm still trying to
figure out
why the hell anyone would want to have "edge" routers (instead of dumb
TDMs) if not for inability of IOS to support large numbers of virtual
interfaces. Same story goes for "clusters" of backbone routers.

When ANY router becomes as reliable as a dumb TDM device, then maybe we can
begin collapsing the POP topology. However, the very nature of the Internet
almost prevents this reliability from being achieved (having a shared
control and data plane seems to be the primary culprit). There are routers
out there today that can single-handedly replace entire POPs at a fraction
of the rack, power, and operational cost. Hasn't happened, tho.

I don't like wasting ports for redundant n^2 or log(n^2) interconnect
either, but router and reliability mix like oil and water...

My 2c.

-chris

2

>Regarding the diameter of the Internet - I'm still trying to
>figure out
>why the hell anyone would want to have "edge" routers (instead of dumb
>TDMs) if not for inability of IOS to support large numbers of virtual
>interfaces. Same story goes for "clusters" of backbone routers.

When ANY router becomes as reliable as a dumb TDM device, then maybe we can
begin collapsing the POP topology. However, the very nature of the Internet
almost prevents this reliability from being achieved (having a shared
control and data plane seems to be the primary culprit).

Uhm. Actually, control & data planes are rather separate inside modern
routers. What is flaky is router software. That's what you get when your
router vendor sells you 1001 way of screwing up your routing :slight_smile:

There are routers out there today that can single-handedly replace
entire POPs at a fraction of the rack, power, and operational cost.
Hasn't happened, tho.

I know two boxes like that - one is broken-as-designed, with copper
distributed fabric; another (courtesy of VCs who managed to lose nearly
entire engineering team mid-way but hired a bunch of marketers long before
there was anything to ship) is still in beta.

I don't like wasting ports for redundant n^2 or log(n^2) interconnect
either, but router and reliability mix like oil and water...

Actually, not. A router is a hell of a lot simpler than a Class-5 switch,
particularly if you don't do ATM, FR, X.25, MPLS, QoS, multicast, IPv6,
blah, blah, blah.

Demonstrably (proof by existence), those switches can be made reasonably
reliable. So can be routers. It's the fabled computer tech culture of "be
crappy, ship fast, pile features sky high, test after you ship" aka OFRV's
Micro$oft envy, which is the root evil.

--vadim

3

now *that* i agree with, and to be even more specific, it's not that any of us are against making profits... it's that many of these vendors, and service providers, have decided to make a profit at the expense of *everything* else (good service, happy customer, etc).

i truly wonder how low the economy, and specifically this industry, is going to have to go before this paradigm is shifted.

-b

4

Thus spake "Vadim Antonov" <avg@exigengroup.com>

Actually, not. A router is a hell of a lot simpler than a Class-5
switch, particularly if you don't do ATM, FR, X.25, MPLS,
QoS, multicast, IPv6, blah, blah, blah.

The data plane is remarkably easier. The control plane is arguable. And
without ATM, FR, MPLS, QOS, multicast, etc. nobody will be buying your router.

Demonstrably (proof by existence), those switches can be
made reasonably reliable. So can be routers. It's the fabled
computer tech culture of "be crappy, ship fast, pile features
sky high, test after you ship" aka OFRV's Micro$oft envy,
which is the root evil.

The question is actually whether anyone would pay the cost of a perfect router.
People complain that today's routers are too expensive, and most vendors are
going bankrupt or giving up. Many of those were marketing to the "featureless
and reliable" niche.

S

5

#include <not-emplyers-words.h>

  I have seen a shift in the past year away from feature, Feature,
FEATURE! and on increasing the quality of software in some router
vendors. they have people that honestly do care, but i've seen them
be thwarted by people who seem to be less focused on providing anywhere
near the number of 9's that their marketing people have promised.

  the question remains will this be another short-lived focus on
quality over quantity or will they really learn from their mistakes
and teach each other within the company before they leave for the
next-big-thing how much this really matters to some of us. i am
worried that they will regress after a period of time should their
focus waver on the ultimate goal.

  hopefully this industry will not end up like the cellular one where
people jump providers to the cheapest rate plan next time their
contract expires or becomes close to coming up for renewal.

  - jared

6

Thus spake "Vadim Antonov" <avg@exigengroup.com>
> Actually, not. A router is a hell of a lot simpler than a Class-5
> switch, particularly if you don't do ATM, FR, X.25, MPLS,
> QoS, multicast, IPv6, blah, blah, blah.

The data plane is remarkably easier. The control plane is arguable.

That's synchronous, jitter-controlled data plane, not the asynchronous,
potentially lossy data plane as in routers. The difference
complexity-wise is significant. That's why people are able to get much
higher speeds in packet switches than in voice switches.

As for the control plane... I also thought call control to be easy, until
I started to learn about it as a part of my job at Genesys Labs (the
leading CTI vendor). The seemingly small number of features (call
progress, hold, forwarding, one-step and two-step transfers, conference
calls, predictive dialing, etc, etc, etc) combine exponentially to produce
huge number of states and possible race conditions. Add to that strong
real-time constraints imposed (foolishly, perhaps) by signaling system
protocols, the resource allocation (as opposed to best-effort behaviour)
the real fault-tolerance, accurate accounting requirements, interfacing
with large-scale CMS, and you get the picture.

And without ATM, FR, MPLS, QOS, multicast, etc. nobody will be buying
your router.

Yep. Because of the need to integrate with all kinds of OFRV powered
feature-rich networks, not because of any rational need to have those
features in the first place.

Microsoft did that to software. Cisco is doing that to networking.
Being a cynic, I do own some Cisco stock, and wait for Mr. Chambers to
figure out that producing boxes which won't break in few years erodes
company's profits by forcing it to compete against its own old models.
Quite a few people are quite happy with CGS-es on their T-1s :slight_smile: eBAY
prices on perfectly good routers are great, too [tongue firmly in cheek].

The question is actually whether anyone would pay the cost of a
perfect router. People complain that today's routers are too
expensive, and most vendors are going bankrupt or giving up. Many of
those were marketing to the "featureless and reliable" niche.

I'm not aware of any whose marketing wasn't focused on the feature
checklist. Not in carrier or enterprise space. In SOHO segment Linksys is
making a killing with their cheap boxes.

Note that "perfect router" to me is the one which you plug in and spend 30
seconds configuring, after which it just works. Your mileage may vary,
but I strongly suspect that a bookshelf full of manuals is not a
desireable attribute of a perfect anything.

--vadim