Friday, August 24, 2007

The Little Engine That Could: How Linux is Inadvertently Poised to Remake the Telephone and Internet Markets

One of the cheapest Linux computers you can buy brand new (not at a garage sale) is the Linksys WRT54G, an 802.11g wireless access point and router that includes a four-port 10/100 Ethernet switch and can be bought for as little as $69.99 according to Froogle. That's a heck of a deal for a little box that performs all those functions, but a look inside is even more amazing. There you'll find a 200 MHz MIPS processor and either 16 or 32 megs of DRAM and four or eight megs of flash RAM -- more computing power than I needed 10 years ago to run a local Internet Service Provider with several hundred customers. But since the operating system is Linux and since Linksys has respected the Linux GPL by publishing all the source code for anyone to download for free, the WRT54G is a lot more than just a wireless router. It is a disruptive technology.

A disruptive technology is any new gizmo that puts an end to the good life for technologies that preceded it. Personal computers were disruptive, toppling mainframes from their throne. Yes, mainframe computers are still being sold, but IBM today sells about $4 billion worth of them per year compared to more than three times that amount a decade ago. Take inflation into account, and mainframe sales look even worse. Cellular telephones are a disruptive technology, putting a serious hurt on the 125 year-old hard-wired phone system. For the first time in telephone history, the U.S. is each year using fewer telephone numbers than it did the year before as people scrap their fixed phones for mobile ones and give up their fax lines in favor of Internet file attachments. Ah yes, the Internet is itself a disruptive technology, and where we'll see the WRT54G and its brethren shortly begin to have startling impact.

You see, it isn't what the WRT54G does that matters, but what it CAN do when reprogrammed with a different version of Linux with different capabilities.

Yes, smartypants, I know that other wireless access points and routers run Linux or can be made to run Linux. It didn't take long for hackers to figure out that Apple's original AirPort access point used a version of the 486 processor and could be convinced to speak Linux. But the WRT54G is different. This is a $70, not a $299 box and its use of Linux is no secret. Linksys, now owned by Cisco, not only doesn't mind your hacking the box, they are including some of those hacks in their revised firmware.

We're not in Kansas anymore.

Probably the most popular third-party firmware you can get for the WRT54G comes from Sveasoft, a Swedish mobile phone software company. Actually, Sveasoft is only kinda-sorta Swedish since the head techie (and for all I know the company's only employee) is James Ewing, a former contract programmer from California. Ewing took time off to visit Honduras where he met a woman from Sweden, and a decade ago moved with her back to Scandinavia, where they live three kilometers from the mainland on an island without broadband Internet service. Looking for a cheap wireless connection much like the one I had a few years ago in Santa Rosa, Ewing discovered through the Seattle Wireless Group web site the amazingly adaptable WRT54G, and has devoted much of his time since to improving the little box's firmware.

If you have a WRT54G, here's what you can use it for after less than an hour's work. You get all the original Linksys functions plus SSH, Wonder Shaper, L7 regexp iptables filtering, frottle, parprouted, the latest Busybox utilities, several custom modifications to DHCP and dnsmasq, a PPTP server, static DHCP address mapping, OSPF routing, external logging, as well as support for client, ad hoc, AP, and WDS wireless modes.

If that last paragraph meant nothing at all to you, look at it this way: the WRT54G with Sveasoft firmware is all you need to become your cul de sac's wireless ISP. Going further, if a bunch of your friends in town had similarly configured WRT54Gs, they could seamlessly work together and put out of business your local telephone company.

That's what I mean by a disruptive technology.

The parts of this package I like best are Wonder Shaper and Frottle. Wonder Shaper is a traffic-shaping utility that does a very intelligent job of prioritizing packets to dramatically improve the usability of almost any broadband connection. If you supposedly have all this bandwidth, but uploading slows your downloading to a crawl or web surfing makes your VoIP phone calls break up, you need Wonder Shaper. At the expense of the top 10 percent of upstream and downstream bandwidth, Wonder Shaper makes brilliant use of what's left over. The result is that not only are phone calls and web serving unaffected by each other but your wireless ISP customers won't have a measurable effect on your surfing, either.

Frottle is another Open Source product, this one coming from a network of wireless networks in Western Australia. Frottle's job is to cure the hidden node problem that was left unsolved in the original Wireless Distribution System (WDS) 802.11 specification from 1999. Hidden nodes are wireless clients or access points that are out of range from one party in a client-AP data transfer. 802.11's CSMA/CD technology assumes that all parties can listen on the line and avoid collisions. But on a wireless network this isn't always possible, so Frottle uses a token-passing scheme (yes, just like Arcnet or Token Ring) to make sure only one node at a time can talk whether the clients can hear each other or not. Maximum bandwidth is limited but maximum throughput is increased, which is why IBM used to argue that Token ring's four megabits-per-second was more bandwidth than Ethernet's 10 megabits.

Neither Wonder Shaper nor Frottle are the most elegant solutions, but they work well and they work together on the Sveasoft firmware.

The result is a box you connect to power, to a DSL or cable modem and MAYBE to your PC (if all you want to be is a service provider the PC isn't needed) and it automatically attaches itself to an OSPF mesh network that is self-configuring. In practical terms, this mesh network, which allows distant clients to reach edge nodes by hopping through other clients en route, is limited to a maximum of three hops as the WiFi radios switch madly back and forth between sending and repeating modes. If you need to go further, switch to higher-gain antennas or gang two WRT54Gs together. Either way, according to Ewing, his tests in Sweden indicate that if 16 percent of the nodes are edge nodes (wireless routers with DSL or cable modem Internet connections), they can provide comparable broadband service to the other 84 percent who aren't otherwise connected to the Net.

There is an obvious business opportunity here, especially for VoIP providers like Vonage, Packet8 and their growing number of competitors. If I was running a VoIP company ,I'd find a way to sell my service through all these new Wireless ISPs. The typical neighborhood WISP doesn't really want to DO anything beyond keeping the router plugged-in and the bills paid, so I as a VoIP vendor would offer a bundled phone-Internet service for, say, $30 per month. I handle the phone part, do all the billing and split the gross sales with the WISP based the traffic on his router or routers. If one of my users walks around with a WiFi cordless phone, roaming from router to router, it doesn't matter since my IP-based accounting system will simply adjust the payments as needed.

The result is a system with economics with which a traditional local phone company simply can't compete.

That's just one idea how these little routers might be used. The actual killer app will probably be something altogether different, but I am convinced this is the platform that will enable it. And that's because what we are talking about here isn't just what you can do with a WRT54G, but what you will soon be able to do with almost any wireless access point.

The cat is out of the bag. This same firmware runs on Belkin, ASUS, and Buffalotech routers today. The source code comes from Broadcom, not Linksys. Linksys paid a Taiwanese company called Cybertan to customize the Broadcom standard Linux distribution that is given to all manufacturers. Two years from now, the current crop of name-brand routers will give way to dirt cheap generics from China and Taiwan with exactly the same hardware and chips. If you look inside the current 802.11g crop from the big names you have basically two routers -- Broadcom and Atheros. They are all based on reference designs and are essentially identical internally.

A well-funded VoIP company like Vonage could today start WISP-based deployment one city at a time. With newspaper ads and direct mail, they could recruit what would be essentially micro-franchisees, each of which would get at cost a pre-configured router (or my preference -- a pair of routers) and a DSL or cable broadband account. Since each node costs the VoIP provider exactly nothing, the problem of flaky franchisees is eliminated by over-building the network and conscientious franchisees make more money as a result. For $50 down and $30 per month the franchisee makes $93.75 per month (provided they keep the connection up and running). Want more revenue? Put routers in all your stores or delivery trucks or in the homes of your friends in exchange for giving them free Internet and/or phone service. Your take per router drops to $78.75 but your gross profit margins are still more than 70 percent.

Or imagine a school or a church distributing routers among parents or parishioners as a fund-raiser. Let's see how long SBC or Verizon lasts against the Baptists. Now THAT's disruptive.


http://www.pbs.org/cringely/pulpit/2004/pulpit_20040527_000456.html