Understanding the Level 3-Comcast spat (FAQ)

Internet backbone provider Level 3 says that cable giant Comcast is threatening the openness of the Internet and Comcast is accusing Level 3 of trying to get a free ride on its network.

Which one is right?

Accusations between the two companies have been flying since the end of the day Monday. Level 3 has accused Comcast of violating Net neutrality principles by insisting that Level 3 pay for sending more traffic over Comcast's network. Level 3 has recently won a contract to deliver video services for Netflix, which is the largest single producer of Internet traffic on the Web in North America.

Comcast argues that Level 3 is trying to use its free peering relationship with Comcast to add twice as much traffic onto Comcast's network without paying for it. Meanwhile, Level 3's content delivery network competitors are forced to pay for the same access Level 3 wants to get for free.

Much of the dispute is focused on business relationships, largely hidden from the average consumer. To help people better understand what's going on, CNET has put together this FAQ.

What is a content delivery network?
A content delivery network or content distribution network (CDN) is a system of servers that is placed throughout the Internet to bring content and data closer to end users. By storing data closer to consumers, Web pages can load faster and streaming movies can be transferred faster than if the data was stored in a central location.

The problem CDNs solve is that they maximize the speed at which content can be delivered into people's homes. CDNs also cut down on the amount of traffic that must travel across Internet backbone networks. This reduces bandwidth costs and saves money while also creating a better experience for end users.

CDN providers, such as Akamai, Limelight, and Amazon CloudFront, have created businesses that send significantly more traffic over another carrier's infrastructure than they receive on their own network. For that reason, they typically purchase services from Internet backbone providers. A source close to Comcast confirmed that Comcast had a commercial arrangement with Akamai to deliver all of its services to Comcast customers, including the Netflix content.

Comcast said in its statement that the dispute with Level 3 is over a peering relationship. What does that mean?
Comcast and Level 3 are services providers who each own their own infrastructure. Comcast is one of the largest broadband providers in the U.S. It connects thousands of homes and businesses directly to the Internet via its local and regional network.

Level 3 is what is called a backbone Internet provider. Traffic from regional and local networks, such as Comcast's broadband network, are aggregated at various points and then connected to Level 3's network. Think of it like the road system in the U.S. Level 3 is a big interstate highway that carries data long distances. At some point, the traffic must leave the interstate highway and get on state and local highways to reach its final destination.

Carriers charge for access to their networks. But if two service providers each own infrastructure, they can create a peering arrangement. If they each carry roughly the same amount of traffic from the other carrier on their network, they can simply swap the traffic for free.

Does money ever get exchanged for these transactions or do they swap the traffic for free?
It depends on the terms of the deal and the nature of the relationship. Networks may charge a fee for using their network, but if each provider is sending equal amounts of traffic onto another carrier's network, then they may simply swap traffic without money changing hands at all.

But when one network sends more traffic to another network, the sender usually pays a metered fee for the traffic it has sent. This makes sense because when more traffic is sent to a network, that network operator must upgrade its network to handle the additional traffic.

Is this the type of relationship that Level 3 and Comcast have had?
Yes, Comcast said in a letter to the FCC today that the amount of traffic the companies sent to each other was roughly equal. Comcast said that Level 3 typically send more traffic to Comcast than Comcast sent to Level 3, but under traditional peering arrangements this was considered roughly equal. And it made sense, since most consumers download more content and media over their broadband connections than they upload.

So is Level 3 a backbone provider or a CDN provider?
The answer is that it's both. Level 3 is one of the largest Internet backbone providers in the world, but it also runs a content delivery network business. Because it owns its own nationwide backbone network, it has an advantage over its competitors, such as Akamai, which do not own any network infrastructure. The big advantage is that Level 3 doesn't have to pay to access its own nationwide backbone that carries traffic to regional networks, such as Comcast's network. But Level 3 does not own the local and regional broadband networks.

Is this where the dispute with Comcast comes into play?
Yes, it is. This is the heart of the disagreement. In that letter Comcast sent to the FCC today, it said that under its former agreement, Comcast and Level 3 exchanged all on-network Internet traffic for free. But Comcast actually paid Level 3 to interconnect its network. Even though the parties exchanged traffic on a 2:1 basis in which Comcast actually terminated twice as much traffic from Level 3 as the company carried from Comcast, the companies considered it an even swap.

Comcast said in its statement yesterday that Level 3's deal with Netflix will double the amount of traffic that Level 3 will send to Comcast's network. The ratio of traffic will soon be 5:1, in which Netflix sends a greater amount of traffic over Comcast's network than Comcast sends over Level 3's network.

What's more, Comcast had been charging Akamai, which previously delivered Netflix's video content. Now that Level 3 has won that business, Comcast is not collecting that fee from Akamai. As such, Comcast argues that it must charge Level 3 the fee to deliver the traffic.

So Level 3 doesn't want to pay Comcast to deliver its traffic even though its competitor, Akamai, was charged for delivering the Netflix content?
That is correct. Level 3 likely won the deal with Netflix because it was able to undercut its competition in terms of pricing. And it feels that since it is a network peering partner with Comcast that it should not have to alter its agreement, even though it is adding a new CDN customer that will consume a lot more bandwidth than other customers using its CDN network.

What does all of this have to do with Net neutrality?
Level 3 claims that Comcast is singling out the Netflix video traffic and attempting to charge a fee for delivering a specific type of traffic to customers on its network. The company argues that it has no other option but to agree to this fee if it wants to deliver Netflix video to customers in Comcast's territory. And so Level 3 claims this is a violation of the Net neutrality principles that the FCC has already established and is currently working to make official regulation. These principles state that network operators should not slow or degrade Web traffic from a competitor in an effort to make its own service more appealing to consumers. Net neutrality supporters have also been pushing the FCC to adopt provisions that would prohibit network operators from charging companies higher fees to deliver high bandwidth content at a higher quality.

What do you think? Is Comcast violating Net neutrality principles or is this merely a peering dispute?
My gut tells me this is more of a peering dispute than a true violation of Net neutrality. And there are several reasons why I think this.

For one, Comcast isn't stupid. I doubt very much Comcast would risk rattling the FCC's cage as it puts the finishing touches on new Net neutrality regulations or as it deliberates over conditions on Comcast's merger with NBC Universal, both of which are happening right now.

Second, Comcast has said that it doesn't care if Level 3 is delivering video from Netflix or high-capacity files from NASA, the fact that Level 3 will be more than doubling the amount of traffic it dumps onto Comcast's network is the problem. Comcast summed it up this way, again in today's letter to the FCC:

"Level 3 is trying to game the process of peering--one that has worked well and consensually, without government interference, for over a decade--in order to gain a unique and unfair advantage for its own expanding CDN service. Level 3's problem apparently arises out of the fact that it recently won a bid to become one of Netflix's primary CDN providers--in competition with the major national CDNs that already send Netflix and other traffic to Comcast's network. In order to undercut its CDN competitors, Level 3 wants to avoid the commercial arrangements other CDN companies use to terminate traffic onto Comcast's and other providers' networks, and instead force Comcast to accept its CDN traffic for free, under a 'peering' relationship. This is not how peering works, here or anywhere in the world. What Level 3 is suddenly pushing--a "new theory" of peering--would throw the traditional, "balanced traffic" peering rulebook out the window, give Level 3 an unfair cost advantage over its competitors, and shift all of the costs from Level 3 and its content customers onto Comcast and its high-speed Internet customers."

Lastly, I find Level 3's assertions somewhat hypocritical. In a press release issued five years ago when Level 3 was in a peering dispute with Cogent, it argued, as Comcast has, for fair and balanced free peering. In that statement, Sureel Choksi, executive vice president of Level 3, said "In order for free peering to be fair to both parties, the cost and benefit that parties contribute and receive should be roughly the same."

Level 3 said that there were a number of factors that determined whether a peering relationship is mutually beneficial. And it determined that the arrangement with Cogent was not fair because it sent far more traffic over Level 3's network than Level 3 sent over Cogent's network.

"It is important to keep in mind that traffic received by Level 3 in a peering relationship must be moved across Level 3's network at considerable expense. Simply put, this means that, without paying, Cogent was using far more of Level 3's network, far more of the time, than the reverse. Following our review, we decided that it was unfair for us to be subsidizing Cogent's business."

It sounds to me like Level 3 is asking Comcast to do something it has refused to do for its peering partners.

So what happens next? Will Comcast customers not be able to watch Netflix movies?
No, Comcast is not blocking Netflix. And it has no intention of blocking Netflix.

"Any rumors about blocking Netflix are false," a Comcast spokeswoman said in an e-mail. "Our customers can and do watch video from any online video provider, including Netflix and dozens of others, on our high-speed Internet service."

FCC Chairman Julius Genachowski said during a press conference today that he will look into Level 3's claims. As I mentioned before, Comcast has already written a letter to the FCC explaining its position. Level 3 has not lodged any formal complaint at the FCC.

Smaller peering disputes are usually settled in court. And the two parties in dispute usually cut off access to each other. But that's unlikely to happen in this case. Level 3 and Comcast are too dependent on each other for sending massive amounts of Internet traffic for their customers. It would make them both look bad if they refused to provide service to each other. It's more likely the companies will settle their dispute between now and January when the bulk of Netflix's traffic is transferred to the Level 3 network.

Demilitarizing cybersecurity (Q&A)

As if the wars on terror and drugs weren't keeping U.S. officials busy enough, the drum beats of cyberwar are increasing.

There were the online espionage attacks Google said originated in China. Several mysterious activities with Internet traffic related to China. The Stuxnet worm that experts say possibly targeted Iranian nuclear centrifuges. An attack on the WikiLeaks site after it released classified documents damaging to U.S. foreign policy. And don't forget the Internet attack on Estonia from a few years ago.

To deal with the geopolitical dramas that are projected in the online world, the U.S. is using military strategy and mindset to approach cybersecurity, creating a Cyber Command and putting oversight for national cybersecurity under the auspices of the Department of Defense.

But offense isn't always the best defense, and it never is when it comes to Internet security, says Gary McGraw, author and chief technology officer at security consultancy Cigital. More secure software, not cyber warriors, is needed to protect networks and online data, he writes in a recent article, "Cyber Warmongering and Influence Peddling."

CNET talked with McGraw about how the militarization of cybersecurity draws attention from serious threats.

Q: So, Tell me what's wrong with going to DEFCON 1 in cyberspace now?
McGraw: I wrote an article with Ivan Arce, the founder and chief technology officer of Core Security Technologies. He's from Argentina. Every time I talk to him he asks 'what is up with you Americans and cyberwar anyway? Why are you so obsessed with cyberwar?' Because nobody else is talking about it in the rest of the world. I travel a lot internationally and he is right. So we started talking about why that was. One of our main points is that there is a confusing blend of cyberwar stuff, cyber-espionage stuff and cybercrime stuff, and the stories are used to justify whatever political or economic end people may have, instead of trying to disambiguate these three things and talk about what they actually are.

What's the danger with that?
McGraw: The danger is that if we lump everything under 'cyberwar,' then our natural propensity in the United States is to allow the Defense Department to deal with it. The DoD set up a Cyber Command in May. Cyber Command has an overemphasis on offense, on creating cyber-sharpshooters and exploiting systems more quickly than the enemy can exploit them. I don't think that's smart at all. I liken it to the world living in glass houses and Cyber Command is about figuring out ways to throw rocks more accurately and quickly inside of the glass house. We would all be better suited trying to think about our dependence on these systems that are riddled with defects and trying to eliminate the defects, instead.

Is the rhetoric all driven by attracting money? That's a very cynical way of thinking.
McGraw: A lot of people think it is. The military industrial complex in the U.S. is certainly tied very closely to the commercial security industry. That is not surprising, nor is it that bad. The problem is the commercial security industry is only now getting around to understanding security engineering and software security. The emphasis over the past years has been on trying to block the bad people with a firewall and that has failed. The new paradigm is trying to build stuff that's not broken in the first place. That's the right way to go. If we want to work on cybercrime and espionage and war, to solve all three problems at once, the one answer is to build better systems.

You mention that cybercrime and cyber-espionage are more important than cyberwar. Why is that?
McGraw: Because there is a lot of crime, less espionage, and very little cyberwar. (chuckles) And the root cause for capability in all these things is the same. That is dependence on systems that are riddled with security defects. We can address all three of those problems. The most important is cybercrime, which is costing us the most money right now. Here's another way to think about it: everyone is talking about the WikiLeaks stuff, and the impact the latest (confidential files) release is having on foreign policy in the U.S. The question is, would offensive capability for cyberwar help us solve the WikiLeaks problem? The answer is obvious. No. Would an offensive cyberwar capability have helped us solve the Aurora problem where Google's intellectual property got sucked down by the Chinese? The answer is no. What would have helped address those two problems? The answer is defense. That is building stuff properly. Software security. Thinking about things like why on earth would a private (officer) need access to classified diplomatic cables on the SIPRNET (Secret IP Router Network)? Why? If we thought about constructing that system properly and providing access only to those who need it, then things would be much better off.

The term "cyber" makes it seem more scary. We're just talking about Internet, right? Might there be a problem with semantics?
McGraw: There could be. There has been an over emphasis on cyber war in the U.S. The problem with cybersecurity is that there is just as much myth and FUD and hyperbole as there are real stories. It's difficult for policy makers and CEOs and the public to figure out what to believe because the hype has been so great, such as with the Estonia denial-of-service attack from 2007. So that when we talk about Stuxnet it gets dismissed.

So it's the boy who cried wolf problem?

McGraw: Yes.
Stuxnet is real. Is that cyberwar?

McGraw: It seems like a cyberweapon. I think it qualifies as a cyberwar action. My own qualification is that a cyberattack needs to have kinetic impact. That means something physical goes wrong. Stuxnet malicious code did what it could to ruin physical systems in Iran that were controlling centrifuges or that were in fact centrifuges. If you look at the number of centrifuges operating in Iran you see some big drops that are hard to explain. (Iranian President Mahmoud) Ahmadinejad admitted there was a cyberattack on the centrifuges.

So why does the attack on Estonia not qualify?
McGraw: The kinetic impact is important, but also an act of war is the act of a nation-state. The Estonia attacks fail the nation-state actor test. It also fails the real impact test. Sure, their network went down, but whoop dee do! Who cares? If you took that same sort of attack against Google or Amazon they wouldn't even notice. I think people were using that attack--which was carried out by individual cybercriminals in Russia, not by the state--to hype up the cyber war thing. In fact, in my work in Washington [D.C.], the Estonia story keeps coming up, over and over again, as an example of cyberwar.

What is your qualification to discuss cyberwar matters and policy?
McGraw: This year, I've been working more in Washington than I have in past. I've been to the White House, the Pentagon, talked to think tanks. I'm a little bit worried that the discourse is too much about cyberwar. We should try to untangle the war, espionage, and crime aspects and maybe emphasize building better systems and getting ourselves out of the glass house as opposed to trying make a whole new cadre of cyber-sharpshooters as [CIA Director] General Hayden suggests. For policymakers the conception of our field [of security] is muddled.

I'm worried we're not spending on [Internet security] defense at all. There's no way to divide and conquer networks. That is, we can't defend the military network or the SIPRNET but not defend the Internet because we're ignoring 90 percent of the risk. Most of the infrastructure in the U.S., 90 percent of it that's important, is controlled by corporations and private concerns, not by the government. The notion that we can protect military networks and not the rest of it just doesn't make any sense. That's one problem. The other problem is the Air Force has always been about domination in the air and taking away that capability from the enemy early and eradicating infrastructure. This notion of a 'no-fly zone' is kind of interesting. Unfortunately those tactics don't work in cyberspace because there is a completely different physics there. There is no such thing as taking ground or controlling air space in cyberspace. Things move at superhuman speed in cyberspace. So some of these guys who are good military tacticians are having a hard time with cyberwar policy and cyberdefense because of the analogies they're using.

You mentioned in your article that "in the end, somebody must pay for broken security and somebody must reward good security." Are you suggesting that we hold software makers liable for flaws?
McGraw: I don't know what the answer is. We need to change the discourse to be around how do we incentivize people to build better systems that are more secure and how do we disincentive building of insecure systems that are riddled with risk? As long as we can have that conversation then policy makers might be able to come up with right sort of levers to cause things to move in the right direction. We're not suggesting any particular approaches, like liability. We're just trying to change the discourse from being about war to being about security engineering.

Anything else?
McGraw: I think we are at risk and I do think cyberwar is a real problem we have to grapple with. But even though we are at risk, we need to have rational conversations about this. Too much FUD and hyperbole don't do anything to help the situation. The poor guys that are charged with setting policy have a hard time doing that because we're having the wrong conversation at the policy level right now.

IBM chips: Let there be light signals

0diggsdigg

IBM's new processors integrate optical communication technology in a development called silicon photonics.

(Credit: IBM)

IBM has achieved a major milestone in making the dream of silicon photonics, in which computer chips send signals of light rather than electricity, into reality.

At the semiconductor industry conference Semicon in Tokyo today, IBM photonics leader Yurii Vlasov is detailing how IBM has created a chip that integrates many of the necessary elements of optical communication between a processor and other devices. Significantly, the design uses conventional rather than exotic chip manufacturing technology, involves very small components, and essentially permits a fiber-optic communication line to be attached directly to a processor.

And more significantly, it's headed for real-world use, a sign that IBM's work is serious. That initial use is in IBM's relatively exotic Exascale project to build a computer that can perform a quintillion mathematical calculations per second--roughly 1,000 times that of today's fastest supercomputers.

"In three to five years, silicon photonics will be the main enabler for that level of computation," said Solomon Assefa, an IBM research scientist and one of the members of the team that developed the chip. And in the years after that, it'll follow the traditional computing industry trend and spread to more ordinary products, he predicted.

IBM's project is called SNIPER--short for "silicon nanoscale integrated photonic and electronic transceiver"--but the company is labeling it with the slightly less ungainly term of silicon nanophotonics. It's been under way for a decade, an indicator of just how difficult it is to develop.

The technology is a breakthrough and shows Big Blue well ahead of competitors such as Intel and Samsung, said Envisioneering Group research director Richard F. Doherty. Much previous work about optical connections involved much larger devices that were separate from the processor itself, he said.

This image shows how an optical fiber can be connected directly to the chip.

(Credit: IBM)

Specifically, IBM said a transceiver that can send and receive signals measures 0.5 square millimeter, and a single-chip transceiver with a capacity of 1 terabit per second could fit on to a chip measuring 16 square millimeters.

"We expect Intel and Samsung will dust off some optical research and home in on what IBM has now proven," Doherty said.

Assefa is keenly aware of where his competitors stand. "We're many steps ahead of Intel," he said, but one start-up, Luxtera, has silicon photonics products now on the market. In November, it announced silicon photonics links that can transfer data at 25 gigabits per second.

Optical communications, which today are used chiefly for heavy-duty networking tasks such as the core of the Internet, have advantages over the more conventional method of sending electrical signals over wires. For one thing, optical communications use less energy and can span longer distances. For another, sending light down optical fibers enables tremendous data transfer capacity--in part because a technology called wave division multiplexing means that multiple colors of light can be sent simultaneously to pack even more capacity into each communication link.

But optical communication technology is expensive. It's expanding to premium services such as Verizon's Fios and BT's Infinity and to Intel's Light Peak high-end computer connection technology, but costs are still high. What IBM's Sniper project does is bring down the cost by marrying it to the semiconductor chip manufacturing industry.

IBM's technology today consists of a processor with six optical communication links. The chip itself has six transmitters and receivers, each capable of handling eight channels of data through the multiplexing technology. The built in are modulators, which govern how a laser is switched off and on to send signals, though the laser is a separate component. Each modulator can oversee bandwidth of 20 gigabits per second, a major step toward IBM's goal of a chip with an aggregate capacity 50 times that, 1 terabit per second.

For reference, today's conventional Ethernet operates at 1Gbps per second and sometimes at 10Gbps in higher-end servers. A capacity of 1Tbps would be enough to transfer the data of 26 DVDs in a second.

IBM's ultimate goal is a multilayer chip that combines processing, memory, and photonics for optical communication.

(Credit: IBM)

"The limitation to processing now is not really a computation limitation," Assefa said. "The problem is they can't talk to each other fast enough or with higher bandwidth. They waste a lot of power. What you want to do is be able to communicate between processors with much more bandwidth. If you can do that, you can make processors work much faster than they do now."

Building links into silicon makes them cheaper to manufacture and therefore more mainstream. "A single high-performance computing machine will contain a similar number of optical communication channels as currently exist in all parallel optical links worldwide," Vlasov said in his presentation.

The photonics work initially will be used over relatively long distances, connecting separate computers or groups of computers housed in racks. Ultimately, though, IBM expects it to be used within a processor, connecting the independent processing cores, Assefa said.

While IBM has been a pioneer in processor manufacturing and hasn't yielded to more powerful chipmakers such as Intel, its fabrication volume tends to be low by industry standards. It uses those chips in premium products, though, such as its own servers and mainframes, and its research gives it clout when it comes to patent licensing partnerships with rivals. IBM has more than 30 base patents involved with silicon photonics, the company said.

One problem with sending lots of data over electrical wires is that it requires significant electrical power--and much of that is lost to waste heat that holds back processing speeds and requires extensive engineering within computers and data centers to keep things cool. Light-based communication reduces that problem.

"A large part of computer power draw is the high-speed buses and cabling between cards and modules and systems. If even a small number of system go optical, processing efficiency and power savings will both soar," Doherty said.

In addition, high-speed electrical communication produces electromagnetic waves that interfere with neighboring electrical operations, since each tiny wire in a computer can act as an antenna. With optical communications, that interference is reduced, Doherty said, with one benefit being that it's easier to develop low-power chips because there's less need to overpower all the interference.

The next steps for IBM are to focus on efficient, reliable manufacturing of its silicon photonics products and building it into its exascale supercomputer. After that, eventually, it will be a part of a larger industry spreading the light farther.

"The way technology works, usually those [supercomputer] technologies find their way in five to ten years back into the lives of ordinary people," he said. "Down the road i'm sure the tech will find its way into the lives of all of us."