PBS | I, Cringely . July 13, 2006 - The Skype is Falling

The original Cringely frog
mascot, circa 1998.
July 13, 2006
The Skype is Falling:
Even Viral Networks Have to Function in a Real World
By Robert X. Cringely
Last week I responded to Intel‘s investment in the Clearwire Corp. WiMax wireless national ISP by trying to put in perspective what WiMax can and can‘t do. Specifically, it can‘t be a very viable competitor to DSL, cable, or fiber-to-the-home simply because of distance and bandwidth limitations. Some readers didn‘t like this analysis because they felt it was far too U.S.-centric and might serve to discourage WiMax development in countries (specifically Zambia) where there may be more open licensing and power limits and where WiMax has got to be better than nothing at all. And of course readers were right: NOTHING I WROTE LAST WEEK SPECIFICALLY APPLIES TO ZAMBIA. But it also brings up a very good point about the relativity of these technologies and how useful they really are.
Most ISPs (these are U.S. ISPs, by the way, not Zambian ISPs) plan their facilities with the idea that no more than 15 percent of customers are active at any time. If some number significantly above 15 percent attempted to be active, service would suffer. Though through the magic of TCP/IP, suffering in this case means service would slow to a crawl, not end outright. Ending outright is restricted to networks with Quality of Service (QoS), like the telephone network, where the system works fine up to about 10 percent utilization then additional users are rejected, that is unless they are on public safety phone lines, which are supposed to work no matter what. But remember after 9/11 when the public service lines, too, failed, which good old TCP/IP kept the Internet dribbling along. So even QoS is not always what it is cracked up to be.
These 10-15 percent numbers appear again and again in network planning, but remember that in the case of ISPs, just because a maximum of 15 percent of users are online doesn‘t mean they are all sending the same number of bits at the same exact moment. ISPs actually expect bit utilization to mirror user activity so only 10-15 percent of that 10-15 percent is actually sending bits at any given second, which is how ISPs size their actual Internet connection. If broadband users are promised one megabit-per-second, the ISP buys 1-2 percent of that amount of backbone capacity per subscriber. Buying bandwidth wholesale and applying this arbitrage formula is how money is made by ISPs, though it puts some pretty clear bounds on how big their network really is.
The same applies for VoIP vendors like Skype. If you are a Skype user then you are used to seeing on your Skype client interface a real-time read-out of how many people are using the system at that very moment. I just looked and as I write this that number is just over 6.1 million. But what does this number actually mean? It means 6.1 million clients were registered with the system just then, NOT that 6.1 million people were talking. If all 6.1 million Skype users tried to talk at the same time, it would probably bring down the system.
Hey, it isn‘t supposed to work that way! Skype is a peer-to-peer (P2P) network, right? And that means its capacity will expand to handle any number of users. No. Skype uses a technology called "Skype peer-to-peer," which has some definite server involvement and therefore finite scalability. In fact, the number of people who can use -- really, actively use -- Skype at any moment is probably back to that 10-15 percent, which in this case would be 10-15 percent of 6.1 million or a maximum of 900,000 users. That‘s a LOT less than the nearly 200 million registered Skype users and gives us a sense of what eBay got for its $2.6 billion.
Skype‘s server involvement works two ways. First there is the registration server that helps you log-in, tells the world you are available, and facilitates connections, some of which are true peer-to-peer. But a lot of Skype connections aren‘t P2P at all. These connections require some server assist because one side of the conversation or another is hidden behind a Network Address Translation (NAT) firewall. NAT was invented a dozen years ago to help preserve IPv4 addressing and has thrived as a poor man‘s firewall for home networks, but NAT is the bane of P2P systems like Skype that often can‘t see nodes hidden behind NAT firewalls.
The way Skype handles this so-called "NAT traversal" problem is by inserting a server in the middle that can be seen by connections at both ends. This server for Skype is called a "super node" and may well be inside your computer without your knowledge, because Skype super nodes use borrowed bandwidth and processing power. Lucky us.
Skype users who are operating in true peer-to-peer fashion are those whose IP addresses, whether static or dynamic, are readily viewable from anywhere on the Internet. That means no firewall, no Zone Alarm, no Gibson Shields Up!, which is a condition increasingly rare among Internet users. For those Skype users who do sit behind firewalls or use Zone Alarm, they connect through a super node that is visible from both ends of the conversation. Again, the super node has to be unprotected, and it has to have a surplus of bandwidth to handle the conversation relay. This kind of wide open connection is even rarer and there are right now only about 20,000 such super nodes on the Skype system.
Each super node can handle about 10 simultaneous connections for a total of 200,000 connections or 400,000 users. If half of Skype calls have to go through super nodes, that means the actual maximum capacity of the system is less than one million callers.
If Skype were to give up the super nodes and instead use servers based in data centers to do the same job, the cost per active user would be about $0.10 per month, so we are not talking about a lot of money. But for a service that looks for ad revenue of less than a dollar per user per month, even a dime is a lot of money and explains why Skype does things the way it does. It also explains why Skype is madly trying to build its Skype-out and Skype-in fee-based calling services because those will easily boost revenue to where an extra dime per customer won‘t matter. But at present it does.
There is an alternate approach that comes to mind for economically scaling systems like Skype -- simply have all the super nodes be provided by the National Security Agency, which I am sure would be happy to do it.
Skype is just another example of a networking start-up that has to grow or die in its effort to keep ahead of its own network inefficiencies. For more of the same concerning last week‘s column subject, Clearwire, here‘s word from a former Clearwire employee:
"You are absolutely correct regarding Clearwire and bandwidth; individual bandwidth is capped at 1.5 (megabits-per-second) unless your signal is outstanding, and you have to be fairly close to get that speed. I‘m not a field tech, so I lack exact figures, but as a general rule no more than 50 percent of customers on any one tower got signal strength sufficient for 1.5 (mbps) or more."
"To speed buildouts and keeps costs low, Clearwire is not using T1 lines for market backbone. They rely on microwave point-to-point connections that may have multiple hops back to the POP. This is quick, scalable, very easy. But microwave is very susceptible to environmental interference; rain, snow, high wind, etc. all block or misdirect the signal. This is very evident in markets such as Texas and Maui. Some markets call for a very expensive tower crew on a weekly basis."
"Failover? Pardon me, but they don‘t even have a standardized naming convention in place. Engineering tries to build markets as rings for failover, but they aren‘t deciding where in a market the company is focusing, which makes it difficult if not impossible."
"Provisioning is pushing untested markets directly to operations, resulting in high failure rates and many false positive alarms. The load these alarms put on the overloaded monitoring system means alarms in live markets are missed. They refuse to support existing markets above the bare minimum level -- the focus is new markets, so it‘s not going to improve."
Uh-oh. (This is Bob again) Add to this the fallout from Clearwire‘s now-cancelled IPO. Workers who were sticking around and working for submarket wages with the idea that an IPO would make them rich are now re-thinking that position. The IPO was cancelled because Wall Street was reportedly not warming to the idea. So having to protect WiMax as a standard, Intel was forced to throw in $600 million. The Motorola-NextNet part of the deal was structured such that Motorola got NextNet for pennies and put $300 million into Clearwire, which finally paid for $300 million in equipment from NextNet, which could report that money as revenue and look like a going concern.
So it is all a marriage of convenience, but that doesn‘t mean WiMax is dead or bad by any means, though I still think the big winner with WiMax will be Sprint Nextel, not Clearwire. Sprint Nextel is by far the largest holder of U.S. radio frequency licenses useable by WiMax.
Even then, WiMax will be a niche play. Readers this week pointed to the possibility of WiMax meshes being used to totally bypass the telco network, something I wrote about back in 2002. But the problem with that idea is that it requires higher bandwidth at greater distances than WiMax will give us. Remember that the longer-range, higher-power bits of WiMax all require licenses and the unlicensed pieces of WiMax have severe power (and hence distance) limitations. The fact is that if such meshes were going to work we‘d see them popping up right now based on 802.11n, which has comparable specs in every way to unlicensed WiMax. Maybe the people who would build such mesh networks simply didn‘t know these facts.
Now they do.