& cplSiteName &
Comments
Newest First  |  Oldest First  |  Threaded View        ADD A COMMENT
<<   <   Page 2 / 2
arch_1
arch_1
12/5/2012 | 1:42:57 AM
re: Chambers Expects Cisco Dominance
Peter Heywood said:
"Do you think the development of the Internet is now predictable enough to make that assertion?
[...] Is the debate over where intelligence should reside in the network - whether we'll end up with a dumb optical core surrounded by edge routers - has now been settled?

I'm curious about this statement. The Internet provides universal any-to-any IP connectivity. I know how to do this using core routers. I do not know how to do this without using statistical multiplexing of some form in the core. The current core uses statistical multiplexors known as core routers and MPLS label switches. The use of legacy statistical multiplexing techniques ATM and FR) in the core is dying out. The use of alternative statistical multiplexing techniques would require a major (and therefore lengthy) technological shift. The use of circuit-switched techniques would require an even more major technological shift.

Given the above, I know of no way to build an Internet with a "dumb optical core", because a "dumb optical core" cannot buffer packets, and therefore cannot provide statistical multiplexing.

Bufferring must be done in the electrical domain, which implies O-E-O at the switches. The cost difference between the least sophisticated type of statistical mux (say, a PVC-only cell switch) and the most expensive (say, an all-singing all-dancing core IP/MPLS router) affects the cost of the electronics, not the optics, and is very small compared to the cost of the links it connects to.

Tony Li managed to say all of this in a single phrase. "until you have optical buffering, you need core routers." I expanded it simply because many people apparently don't know what he meant.

If none of this makes sense to you, please try to draw a simple network with a dumb optical core and smart edge routers, and then try to scale it. Note that to provide universal connectivity, you need a circuit from each edge to every other edge. Note that as traffic flows vary, you need each edge to connect ot every other edge with its full bandwidth: if you have 1000 edges each with 10Gbps of user traffic,and you want a non-blocking Internet, Each edge will need 999 circuit trunks, and each of these will need an optical interface.

The solution, of course is to separate "dumb" and "optical." I can have a fairly dumb core of scalable statistical multiplexors of any type. The
easiest to install and manage in today's Internet will be scalable core routers with a fairly simple feature set: "dumb." The rest of the smarts can indeed go to the edge.
materialgirl
materialgirl
12/5/2012 | 1:42:54 AM
re: Chambers Expects Cisco Dominance
Peter:
To your first question: "Is the debate over where intelligence should reside in the network - whether we'll end up with a dumb optical core surrounded by edge routers - has now been settled?"

Clearly it is not. However, maximum processing speeds imply a minimum of processing. So, you need to keep as much intelligence out of the core as you can.

However, the clever folks who invented the internet did not anticipate issues like bad guys and QoS needs. The JNPR/NSCN deal, for instance, is interesting in the amount of intellegence that may put into the edge. Storage applications and voice are other spoilers to this "dumb network" apple cart.


To your second question: "Up until now, yesterday's core routers have often been repurposed as today's edge routers. Is this not going to happen in the future?"

This assumes a continuation of Moore's law. If you REALLY talk to leading semiconductor folks, they do NOT assume this is going to keep going. The 90nm node, for instance, is causing skyrocketing NRE and processing costs, in addition to a nasty leakage problem. INTC is moving away from using clock speeds to describe chips in part due to this issue. Something else, like better software, clever multiprocessing or an incrased use of optics, will have to fill in. USer behavior will have to change accordingly.

In fact, all this talk about 500% annual upport games is nuts. Who is going to pay for this? At the end of the day, the network will support paying customers, period. No ROI, no bandwidth. If the price per bit does not continue to plummet, the bandwidth usage will not continue to grow in line with activities with minimal return.
Tony Li
Tony Li
12/5/2012 | 1:42:45 AM
re: Chambers Expects Cisco Dominance

The ones that I have talked to have a vision of many providers' equipment (gives them price leverage) all working together nicely under the same OSS. Every carrier that I have ever talked to (includes Verizon, SBC, Bell Canada, MCI, Worldcom, Qwest, BT, and a few smaller ones) know that their POPs are not the same and therefore cannot use a single device. It just does not make economic sense.


Stephen,

Perhaps we're communicating at cross purposes. Yes, the carriers need to have multiple suppliers, but that need not affect the POP architecture. Some suppliers can be used in one location, some in another. For example, one carrier that I know of uses European ADM's in all of their East Coast POPs and then uses Japanese ADM's in everything west of the Mississippi. They can fluctuate this line to ensure that they get responsiveness from their various suppliers.

Nowhere am I trying to suggest that all POPs are the same. The basic point is that the differentiation between types of routers is wholly artificial and not technical. Once an infrastructure system with an extensible interconnect is installed, you simply populate it with the line cards and media adapters that are necessary for the particular application. Just as the 5E doesn't come in a single fixed configuration, I'm suggesting that the fabric becomes the infrastructure for the POP that all switching is built upon.

Tony
stephenpcooke
stephenpcooke
12/5/2012 | 1:42:42 AM
re: Chambers Expects Cisco Dominance
Tony,

It sounds like you are singing a data-centric version of the 5E/DMS mantra of 20 years ago!

On a somewhat similar vein concerning equipment longevity...

I was intimiately involved in Nortel's first manufacturing divestiture several years ago. I was working for one of the bidders though I had already been a system design authority within Nortel. When it comes to equipment development, the engineers make choices based on the technology available to them today. The market imperatives are simply that they have to get their product out ASAP without too much concern as to what may happen in a few years time.

During the Nortel divestiture exercise, the Nortel 'decision panel' wanted cost reduction suggestions on various pieces of equipment that the manufacturing of which were to be transferred to the winning bidder. One of the pieces of equipment was a double shelf-height DMS peripheral. On closer examination there was a huge wire harness that joined 2 separate backplanes (this was in the year 2000) that had to be manually wire-wrapped onto the backplane pins. The harness was built in Mexico because it was so manually intensive and labor is cheaper there.

The point is that, in 1985 when hardware development on that platform ceased, double height backplanes were very new items and were not trusted. For 15+ years Nortel built frames with this stuff. They never re-designed it (or allowed anyone to do it for them) because they weren't sure if it would work well enough and didn't want to spend the time or money to re-test anything on that platform. In the time following cessation of hardware development on that platform they shipped thousands of these things and designed several other switch fabrics that should have knocked this one out of production.

Bottom line: You never know how long something will last in this business.
russ4br
russ4br
12/5/2012 | 1:42:37 AM
re: Chambers Expects Cisco Dominance
For 15+ years Nortel built frames with this stuff. They never re-designed it (or allowed anyone to do it for them) because they weren't sure if it would work well enough and didn't want to spend the time or money to re-test anything on that platform.


How much has the fixed line subscriber base grown in these 15 years? Was it at least 10X?

The fact that the data traffic is expected to grow orders of magnitude beyond that might make it necessary to come up with new systems design before the 10-20yrs that people are talking about.

- russ
sgan201
sgan201
12/5/2012 | 1:42:34 AM
re: Chambers Expects Cisco Dominance
Hi,
1) Optical Core versus Router Core
In Internet traffic, almost all traffic is not internal and it is send to the Internet. At most, it is dual star configuration. It is not out of question that if in the future, the wave length switch is cheap enough, we may have the situation that the core is optical and we have one arm Core router in a few super pop. The WDM switch will not do any packet processing period.. The full mesh argument in SP's internal Internet network is a flawed argument.

2) Scalable chassis archicteture.
A fabricless design is always cheaper than a chassis with switch fabric.. Now, we can design a fabricless full mesh design at 30Gbps to 40Gbps.. It is not hard to envision a future down in the road for a single chassis fabricless design scale up to Tbps. In that case, how can a multi-chassis multi-racks design at Tbps compete with a single chassis design at both costs and performance..

Dreamer
stephenpcooke
stephenpcooke
12/5/2012 | 1:42:23 AM
re: Chambers Expects Cisco Dominance
Russ wrote:

How much has the fixed line subscriber base grown in these 15 years? Was it at least 10X?

The fact that the data traffic is expected to grow orders of magnitude beyond that might make it necessary to come up with new systems design before the 10-20yrs that people are talking about.


The point was that they HAD made new system designs during that time but the 'tried & true' design, though old and ugly, lived on. There is no way of telling ahead of time what design will stick and what will not.

The data traffic growing orders of magnitude is something that must be taken with significant amounts of salt. Tony's math, though probably mathematically accurate, ignores some basic things like the adoption and deployal rates, the population, their average bandwidth budgets (ie: what bandwidth they are willing to pay money for), the demographics of that population, etc.

During the bubble these same people who are predicting orders of magnitude of growth were stating that the bandwidth requirements in America's networks were doubling every 3-4 months. Where did predictions like those get us?

Russ's choice of words was interesting:

"The fact that the data traffic is expected to grow..." This does not mean that it is a fact that data traffic will grow orders of magnitude, just that it is expected that this will happen. Another poster was entirely correct when they said that there has to be profit in supplying the necessary bandwidth, not just for the equipment providers, but for the carriers. If that profit does not exist due to any reason (eg: competitive price erosion; cannibalization of other, higher revenue services, etc.) adoption rates and service offerings by carriers had better NOT be that high because they will go out of business. How will that help anyone?

Steve.
OneDesignGuru
OneDesignGuru
12/5/2012 | 1:38:05 AM
re: Chambers Expects Cisco Dominance
Love the comments from Wilecoyote, Tony, Li, stephenpcooke, and arch_1. I agree with most of these reponses except the following. "The use of legacy statistical multiplexing techniques ATM and FR) in the core is dying out." I have heard (mainly from Cisco) that ATM and SONET is dying. However, as a major Telecom Gear Maker, we are still seeign heavy requests of devices with ATM and SONET. In my opinion it will take over a decade or two before we see the complete MPLS utilization over ATM mainly because in today's Wireless systems such as Basestations the use of ATM over any other protocol gives predicted channel utilization.
In addition, MR. Chambers is saying that "Once it's in place, we want it to be there one or two decades," The 2 decades of Service Life is a joke to Large Carriers. That is why Nortel, Alcatel, and Lucent all have to provide at least 25 years of service life. I know for fact that the Lucent 5ESS switch was designed to support 40 years of Service Life and according to Telcordia, there is still a utilization of the 1E switch which was built over 40 years ago.
No matter what product will be offered, replacing of the old equipment is not an easy job. In addition, the use of Single suplier equipment will happen "WHEN HELL FREEZES OVER."

Alex.
<<   <   Page 2 / 2


Featured Video
Upcoming Live Events
October 22, 2019, Los Angeles, CA
November 5, 2019, London, England
November 7, 2019, London, UK
November 14, 2019, Maritim Hotel, Berlin
December 3-5, 2019, Vienna, Austria
December 3, 2019, New York, New York
March 16-18, 2020, Embassy Suites, Denver, Colorado
May 18-20, 2020, Irving Convention Center, Dallas, TX
All Upcoming Live Events
Partner Perspectives - content from our sponsors
Sports Venues: Where 5G Brings a Truly Immersive Experience
By Peter Linder, 5G Evangelist, North America, Ericsson
Multiband Microwave Provides High Capacity & High Reliability for 5G Transport
By Don Frey, Principal Analyst, Transport & Routing, Ovum
All Partner Perspectives