Visit Your Local PBS Station PBS Home PBS Home Programs A-Z TV Schedules Watch Video Donate Shop PBS Search PBS
I, Cringely - The Survival of the Nerdiest with Robert X. Cringely
Search I,Cringely:

The Pulpit
The Pulpit

<< [ One Size Fits All ]   |  The Multicast Way  |   [ Prisoner of Redmond ] >>

Weekly Column

The Multicast Way: Why IP Multicast Doesn't Work the Way People Keep Telling Me It Does. Also, Stupid Windows Vista Rumors

Status: [CLOSED]
By Robert X. Cringely
bob@cringely.com

Following my recent pair of columns on emerging peer-to-peer media delivery systems, dozens of readers have responded with the same question: why not multicasting? The way to efficiently use bandwidth for video or audio, they say, is multicast addressing, where one file is sent to hundreds, tens of thousands, even millions of recipients. Next to such capability, p2p seems just plain silly.

But p2p works.

IP multicast has been around as long as we have had an Internet, with an entire class of addresses and ports assigned to their use since the dawn of Internet time. Class D IP addresses occupy one sixteenth of the complete IP address range and are intended solely for multicast use. All Class D addresses begin with 1110 as the first four bits, which means the first octet comprises 224 through 239 and the total address range is 224.0.0.0 through 239.255.255.255. If you are counting, that's 268,435,456 multicast groups. Though the last 28 bits of each Class D address aren't as explicitly set out as it is in some other classes, the RFQs establish three broad ranges of multicast addresses. 224.0.0.0 through 224.0.0.255 is supposedly reserved for "well-known multicast addresses," which to my way of thinking might indicate a continuously operating single channel like, say, BBC1 or maybe HBO. If you wanted to watch whatever happened to be playing at that moment on those networks, then you know where to go. The next range is from 224.0.1.0 to 238.255.255.255, and is intended for Internet-wide or "globally-scoped" addresses, which could be a specific episode of "The Sopranos," whether actually served by HBO or some other network as a rerun. Finally, the address range from 239.0.0.0 through 239.255.255.255 is for local multicast addresses, those on an individual LAN or subnet. Continuing the programming metaphor, these addresses could be used for the location of HBO on your local IP cable or DSL system or for a single "Sopranos episode," again on your local IP cable or DSL system.

Obviously a lot of work was done in this area, so why is it that more than 268 million Class D addresses are going mainly unused? It's because of a mix of rational and irrational fears. The biggest fear is that turning on multicast across the entire Net would end the whole party, bringing the Internet crashing down. This fear, by the way, is probably unfounded. Notice I wrote "probably," because it has never been tested.

The untested problem, as I understand it, has to do with the way Class D addresses are advertised and logged. The etiquette is different than more normal Internet addressing, with Class D using a "come hither" approach of each host declaring, "here is a list of Class D addresses of which I consider myself to be a party, now send in the flicks!" Each multicast-enabled router compiles the Class D lists from all its local hosts, then -- if the addresses are local multicast addresses -- sends on the content from some local server. But if the multicast addresses include some from the Internet-wide or well-known ranges, then each router forwards those addresses to other upstream routers following whatever routing algorithm is specified, with the idea being that if nobody on your subnet wants a particular episode of "Saturday Night Live," then your router will never see those bits. The irrational fear part says that many net admins simply don't believe the system won't practically devolve to something quite different, with global flooding of the network as bits flow whether they are wanted or not and your mother can't watch the news because of some errant and obnoxious copy of "Truth or Dare."

So most net admins simply leave multicast addressing turned off in their routers, where it has been idling for the last several generations. Nearly every router on the Internet is multicast-capable, but hardly any are multicast-enabled.

Now for the rational part of this fear: IP multicast is a big job for a router involving not just this reversed addressing syntax, but also the router's responsibility for delivering the same bits to multiple hosts at the same time with requisite asymmetric retransmissions. That means going two sizes higher when you order your router if you want to broadly use multicast and not suffer performance degradation.

So another reason net admins leave multicast turned off is to save money.

The result is that nearly all multicasting is done locally on LANs or subnets, which is precisely the way it is generally used for DSL-based IP TV systems that are accessing a quite limited range of multicast content sources, generally over a single hop. Broader multicasting can be done using IP tunneling to run the multicast as a unicast through all those interim routers (with their Class D addressing turned off), then ideally going multicast for the last leg, only in nearly every IP tunneling job, there is only one host waiting at the end of the tunnel.

Personally, I don't worry too much about IP multicast. If it were turned on I believe it would work, but I don't expect it to be turned on. If the Mbone had been such a success, maybe we'd see more multicasting, but I don't think the Mbone proved much of anything. Cisco tried like crazy in the late 1990s to promote multicast, for which it bought Precept Software and the original IPTV application just to show the world how television could be done right in an IP-centric world. Only nobody bought the concept or much Precept software. You can learn more about Precept and IPTV in the new episode of NerdTV, which is with Judy Estrin -- who founded Precept then sold it to Cisco and became Cisco's CTO for a couple years. Judy says she built Precept's IPTV around IP Multicast not just because of its obvious efficiency, but because she had read Cisco white papers on IP Multicast and, "Sadly, I believed them."

No, IP multicast is not THE solution to wide deployment of video content over the Internet. It is one possible part of such a solution for ISPs who choose to enable or impose it. But p2p retains the advantage of running well on networks that don't choose to support multicast, which is to say pretty much the whole darned Internet.

Now to this week's announced shipping delay for Windows Vista, which for the previous 20 or so shipping delays was called by its codename Longhorn. I don't know what's up with this, but I can't say I am surprised. Vista is a huge piece of work and Microsoft wants it to be polished to a fine shine before it ships, fails miserably, and has to be extensively patched. This is the company's last chance at denial, so I don't begrudge them at all.

But my readers do.

In dozens of messages on this subject from readers, none were surprised at the shipping delay from November 2006 to January 2007, but many came with theories about ulterior motives for the delay, and especially about why Microsoft would choose to announce it now, rather than later?

As best I can tell it comes down to four basic theories, which, curiously, don't have to be mutually exclusive. You could just as easily apply two or more of these puppies and they'd fit right together like Legos.

Vista Delay Theory Number One: It's the xBox, Stupid. With Sony having just bailed on its own Christmas 2006 introduction for the PlayStation 3, Microsoft wants to make Christmas 2006 solely about its xBox360. Screw the PC makers, they can just sell more boxes in the January quarter. The timing in this case is based solely on Sony's announcement a few days before.

Vista Delay Theory Number Two: Allchin Refuses to Leave the Building. When Vista ships, Windows co-presidente Jim Allchin retires from Microsoft. Maybe Allchin doesn't want to retire just yet. So Vista needs a little more polishing. Of course there is nothing to back this up, yet why does it sound plausible? Why now? Who knows? Who cares?

Vista Delay Theory Number Three: Punish the OEMs. Much of Microsoft power comes down to its near-total control of its hardware OEMs. If you read some of the Microsoft anti-trust materials, there seemed to be a continuing theme of Bill Gates threatening to not renew some big company's OEM license, like he would really shut down Compaq. It was such an effective weapon that it never had to actually be used, but that left a thousand other cuts to inflict just to be sure those OEMs didn't become complacent and start thinking that they ran their own businesses. So why not delay Vista by a couple months just to remind this dwindling group just who is still the sheriff in town? It ruins what would normally be the best quarter of the year, but guarantees the next quarter will set records. Why announce it now? So the companies can reset their production quotas and get ready to help sell xBox360s.

Vista Delay Theory Number Four (my personal favorite): Punish Steve Jobs. Microsoft has found Windows XP runs fine on Apple's new Intel Macs, with sites like GearLog even publishing performance benchmarks for XP on the new MacBook Pro. Worried that the same fate might befall Windows Vista, Microsoft suddenly realizes that the new OS can't possibly be ready for production simply because it CAN run on Apple hardware, just like the good old days of MS-DOS 2.0 when the Redmond insider's motto was supposedly "DOS ain't done 'til Lotus won't run." (See this week's links for a Microsoft insider's debunking of this DOS 2.0 story.)

Chances are none of these theories are valid, but I can't imagine a single reader of this column who doesn't see any of them (and while we're at it why not ALL of them, since they add-up so nicely) as being at least plausible. This includes readers who actually work for Microsoft. That says something terrible about Microsoft's corporate character or at least our perceptions of it. I'm working on another story on Microsoft's (lack of) character that's far bigger and more disturbing than this and loaded with real facts. It might be ready for next week or may still need a bit more work, but it's coming soon. And it will be shocking.

Comments from the Tribe

Status: [CLOSED] read all comments (0)