Rotten at the Core?

BURLINGAME, Calif. -- As massive restructurings hit the service provider world, there is a rising fear that core router upgrades could be put off indefinitely, according to discussions here last week at an NGN Ventures panel on core routing startups.

Startups are particularly worried that once bankrupt carriers have written off lost networking assets and improved their balance sheets, lower bandwidth prices will force the carriers to "sweat out" their existing networks and put off upgrades until their equipment can no longer handle the traffic load (see Report: Core Router Market Falls 22%).

David House, chairman, president, and CEO of Allegro Networks Inc. acknowledged that declining carrier asset values are not helping matters. But even if carriers are now spending less because of this, "they are still spending billions," to handle more data in their networks, he said, "and they'll keep spending where the growth is."

Startups found themselves on the defensive here, wrestling with questions from the audience that reflected a swell of doubt about the near future of the core-routing business.

One skeptical attendee tried to put the panelists on their heels, asking: If carriers can't make money off data services now, why would they want to buy your gear and scale their networks?

"Carriers have always invested up front in a business that they know will be their future," said House. "The idea that [carriers] cannot continue to invest would be death."

Others bullishly insisted that carriers will be forced to pay the price of constantly rising IP traffic.

"You can only sweat assets for so long," said Brian Barry, CEO of Hyperchip Inc. "At some point [service providers] will have to make money from packets."

"When the excess capacity in the network runs out, [Internet] users will not stop creating more traffic," said Larry Roberts, chairman and CTO of Caspian Networks. "There may be some consolidation among backbone carriers, but even with fewer carriers there will still be the same amount of Internet traffic."

"Carriers can make money on IP services, they just don't have the right tools now," added Barry.

Agreeing that carriers will someday buy next-generation routers, the panel admitted their challenge is to survive until such spending takes place. "That's why we raised the money we did," said Roberts, whose company recently recapitalized. "You have to be able to last."

Delays in the startup world raise another question: What does it mean for routing kingpin Cisco Systems Inc. (Nasdaq: CSCO)?

Several discussions about routing focused on limitations of today's core routers, Cisco's specifically. Service providers, it seems, tend to buy a pair of routers for every one needed so that they can build a redundant network.

Roland Acra, VP and general manager of Cisco's Internet Routing Group, replied that "there is a lot of mythology attached [to Cisco] because we were lazy and sloppy as an industry." He said that carriers who bought routers in pairs were focused more on quick growth than reliability.

Acra says despite his competitors' knocks against Cisco, the company's core routers have achieved so-called "five nines" reliability in service provider networks. (And, he said, it's up to Cisco's customers whether to identify themselves.)

He added that startups will have a tough time breaking into the two-horse core router race between Cisco and Juniper Networks Inc. (Nasdaq: JNPR), because they'll need to be skilled in software, hardware, and silicon, all of which can be very expensive to a young company (see Juniper Goes Terabit With the T640). "Admission to the club is a trick, because they have to hit on many cylinders at once."

However, thanks to the evolving data network, incumbent equipment providers may eventually find their installed bases disadvantageous, according to Geoffrey Yang, a general partner at Redpoint Ventures: "An installed base is a double-edged sword. Can you make [old] software do what it wasn't originally designed to do? It could be an issue and might open up an opportunity for startups."

— Phil Harvey, Senior Editor, Light Reading
Page 1 / 9   >   >>
skeptic 12/4/2012 | 10:29:05 PM
re: Rotten at the Core? In reality how big is the core network. Our resident expert newman said that the biggest network is 15 OC48s.
Allowing for redundancy that means max of 30 Oc48s or 75 Gig.

Why is Juniper terabit T640 and other wannabe terabit routers needed?

They are needed because traffic continues to
grow and there is a need for a system which
can grow with that traffic without having to be
replaced after 12-18 months.

There are already providers who have
outgrown the M160/GSR. What people typically
do in such a case is to build an effectively
larger router out of smaller routers. But
that only scales so far and is very expensive
becuase of non-revenue ports introduced into
the system.

Just understand, its not always the case of
needing a terabit router, but as often needing
something that is bigger than what is
conventionally available today and wanting to
have a growth path over time for the equipment
that is of reasonable duration.
lghtswtch 12/4/2012 | 10:29:07 PM
re: Rotten at the Core? Why is Juniper terabit T640 and other wannabe terabit routers needed?

One good, general reason, nothing Juniper in particular: To be able to run the routers at a steady 10-15max percent utilization, avoid much multi-threading and so work around a likely bunch of still-unseen software bugs ;-)
fiber_r_us 12/4/2012 | 10:29:09 PM
re: Rotten at the Core? >In reality how big is the core network. Our
>resident expert newman said that the biggest
>network is 15 OC48s.
>Allowing for redundancy that means max of 30
>Oc48s or 75 Gig.

First, there is no such thing as *the* core network. Each major service provider runs thier own independent national transport backbone. These providers at least include AT&T, Sprint, WCOM, C&W, Level3, Qwest (the so-called "Tier 1" providers). I don't imagine any of the LECs (Bell South, SWB, Verizon, etc) have any interest in big core routers because they are not really serious IP players anyway.

The transport backbone that each service provider maintains has its bandwidth divided into pieces for each of the overlay networks. For most providers, these overlay networks include capacity for voice, ATM, FR (which may be run over ATM), and thier Internet services. Some progressive providers place all of the above services over MPLS :). These latter providers require routers to grow quicker.

For any given provider, the "size" of their "core" network is a difficult thing to measure and describe and thus the size of router required for a given network is just as nebulous. This is mainly due to:

- The "core" is a network of independent links between cities. It is the size of the largest links that most people refer to when they say "my backbone is OC-192".

- Links come in different sizes

- Cities come in different sizes

- Connectivity between different "adjacent" cities in a network's topology varies depending on traffic patterns.

- Network architectures and how protection is accomplished affects the size of routers.

So, if I were to look at a particular city, which had some number of connections to adjacent cities, the "core" routers would be responsible for:

- providing ports from the city it resided in to all adjacent connected cities

- providing ports that connected to equipment *within* the city in which it resided in order to tranport traffic from that city to the rest of the network.

- possibly providing for interconnectivty for the infrastucture within a given city (this depends on your network architecture within that city).

Let take as an example:


- Chicago has links to the following cities with the following capacities (protection included):
-- New York 2xOC192
-- Washington DC 2xOC192
-- Atlanta 2xOC192
-- Dallas 2xOC192
-- Denver 2xOC192
-- San Francisco 2xOC192
-- Detroit 2xOC48
-- Cincinatti 2xOC48
-- St Louis 2xOC48
-- Boston 2xOC48

- Chicago also requires bandwidth local to the city to support peering and customers within the city to the tune of 8xOC192.

- The "core" nodes are two separate routers, each with one of the links from each of cities in the list above.

This means each core router would require (minimally): 10xOC192 ports and 4xOC48 ports just to provide the required connectivity (which is bigger than an M160, but would fit in a 12416 or T640).

This illustration is an example of the size of networks *today* in the Tier 1 providers. The network traffic is growing at 100-300% a year (depending on who you ask). Even at 100% a year, next year this same city would need twice the port capacity on its core routers.
rjmcmahon 12/4/2012 | 10:29:11 PM
re: Rotten at the Core? The flaw in the model may be the exclusive use of bandwidth as the primitive for revenue generation. Oversubscription breaks down in this model. It's like a 24 hour gym being located off an expensive tollway. Nobody could ever make any money.

The new models have to based from connectivity, applications, and content in my opinion. For example, consumers are taxed by monopolists if they want to talk to their relatives so they choose email instead. More specifically, consumers pay $20/month to send/receive email from home. They don't care about how much bw they use, nor should they.

The connectivity providers seem like the next missing piece to the puzzle. They will need to be extremely cost effective and won't be able to afford subsidizing our representative government.

PS. Value creations are exchanged. Money is the secondary means of exchange and has no inherent value. Money is a number stored on paper or metal. (More recently humans started using electrons and photons for its store and its transaction) Money exchange without value exchange is a worthless use of our time.

"Show me the value" seems the better cliche.
yorker 12/4/2012 | 10:29:18 PM
re: Rotten at the Core? In reality how big is the core network. Our resident expert newman said that the biggest network is 15 OC48s.
Allowing for redundancy that means max of 30 Oc48s or 75 Gig.

Why is Juniper terabit T640 and other wannabe terabit routers needed?
Bumper_car 12/4/2012 | 10:29:21 PM
re: Rotten at the Core? Some of these posting have talked around the subject of money. The real problem for all service providers, old and new, is how to be profitable. It takes a lot of money to build and maintain the transmission service infrastructure, before you start putting services on it.

Value added services like voice switching or the Internet are a lot more complex than simple TDM services. Because of that added complexity, they have to generate more revenue per used core bandwidth than do TDM services.

The back office systems, although some remain in Cobol flat file systems (Those old systems don't cost a whole lot to maintain.) A lot of the back office systems are written in C++ or some other language; Oricle is a major software vendor to the telco industry.

The sales people that sell the services have to make thier commissions. The operations support people that keep it running have to be paid. The executives have make their money. The stock investors demand that they make some money as well.

All of this costs the end customer. In order for value added services to have any kind of profit margin, they are implemented with what is called statistical gain. For example, the Internet normally has a 400 to 1 statistical gain between the end user and the backbone. Voice has anywhere between 10 to 1 to 40 to 1 gain. The difference between Internet and voice is that the Internet service gets degraded while at call busy hour, calls get blocked without degrading the voice quality.

It costs a lot of money to upgrade a physical plant or back office software system. Decentralization costs more in labor costs but does provide a better surviability.

The real issue is what does it cost to implement and maintain a service based on a particular technology relative to what customers are willing to pay for it. Right now, TDM is the least expensive for actually used bandwidth of any service. Why do I say that? TDM does not have any statistical gain. If you look at your Internet service, being charged for at a flat rate(primarily because it costs more to bill for usage than it would make) at about $20 per month for a 56k dial up. At 400/1 gain that 56k dial up actually is costing the aggregate of end customers $800 per month. If you check with your local service provider, a TDM 56k line is somewhere around $200-$400 per month, depending on where you are. Yet even at those gain rates, ISPs and NSPs are barely making any profits.

The issue of IP vs TDM is complex and not going to be resolved overnight. I think that over time, more IP services will be able to charge higher rates in order to get the end user bandwidth up and the gain rates down to improve service reliability, availability, etc. In the mean time, many of the legacy free service providers are failing because the services are too expensive to maintain for what the end customers are willing to pay for it.

As someone said "Money talks, everything else walks."
skeptic 12/4/2012 | 10:29:23 PM
re: Rotten at the Core? 3.) The RBOCs funded their networks, built their networks, installed their networks and then the 1996 telcom act tried to steal the networks to give to the poor... (Couldn't help throw in some Robin Hood)

The RBOCs are government-sustained monopolies.
Their core business is protected by government
regulations which as often as not make competition
with them nearly impossible.

The 1996 reform act was supposed to give them
regulatory relief in many areas in exchange for
certain concessions on their part.

If the RBOCs think the 1996 reform act is
resulting in people "stealing" their networks,
then its time to undo the whole 1996 package
including the parts that were huge concessions
to the RBOCs. But thats not going to happen.

As usual, deregulation means the RBOCS get
regulatory relief while ignoring any of the
mandates placed on them in exchange.

If its impossible for the RBOCs to help with
local competition, it should also be impossible
to grant them regulatory relief in other areas
or to allow them entry into new businesses.
lghtswtch 12/4/2012 | 10:29:33 PM
re: Rotten at the Core? johnmoses, try to make a distinction between the OSSs and OSMINE. The latter is a process, a service. One can live with the OSSs and at the same time try to cut/phase out the dependency on OSMINE. Today, Telcordia's position is what it is GÇô tomorrow could be different.
Tony Li 12/4/2012 | 10:29:34 PM
re: Rotten at the Core? The bottom line is you have RBOCs who have been around for Decades, DECADES (thats Multiple) who are servicing the entire population. How do you expect to have them change?

How do they expect not to?

Light-bulb 12/4/2012 | 10:29:36 PM
re: Rotten at the Core? Ok, guys guys how in the world did we start down the Piracy issue again good grief... I thought we left that in "Access Is Gold"
In any event there is lots of text in that article talking about some of these RBOC issues but to restate.
1.) Telcordia is the Model company
2.) The RBOC hate Telcordia too but do they stop paying them? Nope
3.) The RBOCs funded their networks, built their networks, installed their networks and then the 1996 telcom act tried to steal the networks to give to the poor... (Couldn't help throw in some Robin Hood)
4.) The OSMINE process and requirements are NOT going away in the foreseable future Startups need to understand that and continue to put the funding into the business plan.
5.) Why in the world are we arguing on this?
Yes today TIRKS cannot sell more bandwidth than is on the network/trunk side however Telcordia is working with vendors now to develop ways to provision Ethernet services in Increments of VT1.5s. (Yes this Vendor was Smart to make it usable in TIRKS) Even thought its a Layer 1 service the network will be able to support Ethernet connections and handle them in the same fashion as provisioning T1s.
Now fiber I understand your frustration... But just how much "Revenue" did it take to build that stand-alone network? How much of that "1B+" generation is going to pay off the network? And do you support 20Million subscribers?
The bottom line is you have RBOCs who have been around for Decades, DECADES (thats Multiple) who are servicing the entire population. How do you expect to have them change? Some one said it well in identifying all the things that would have to be given away for the RBOCs to even consider. Not too mention that just the Tirks Database itself is the "Largest In the World" Yea, lets give the entire software guys in the lab to the RBOCs to enter Database records into a new system... I'll give you a hint AT&T tried this... Spent 400+ million to develop a better system than moth balled it.
My advice to other vendors in particular... Suck it up, pay Telcordia the 7 Million to get OSMINE and position the product in the LEC (who still have money) don't try to go in sobbing because they won't look at you without OSMINE (Boo Hoo its soo expensive...)
:) And guys please don't let RJ get back into his World Crusade to rid the world of Piracy... Good grief

Page 1 / 9   >   >>
Sign In