Slashdot Mirror

We're still not doing congestion control very well. DOCSIS 3.0, the new cable modem/hub standard, has many congestion management features, but they're a collection of features, not an integrated strategy.

Realistically, there are two QoS options a congestion strategy for general IP-based networks can deliver:

• Low latency, low bandwidth. This is what you want for VoIP and for the low-latency channel of games. For this to work, the network has to enforce the "low bandwidth" requirement by limiting the number of packets in flight. "Fair queueing" can do that on the network side. If you only have one packet in flight (i.e. you wait until each packet is delivered before sending the next one), you shouldn't see any packet loss. If you send more than that, you lose packets. TCP already plays well with fair queueing. UDP-based VoIP protocols that don't do adaptive congestion control need to be fixed.
• High latency, high bandwidth For everything else.

There are fancier reservation schemes, where you can reserve bandwidth, but they only work when all the players in the path cooperate, which tends not to happen. But there's no reason not to get the simple mechanisms above right.

Amazingly there was only one intelligent thing said in the whole article. "Digital switching is key" is correct. Whats amazing is that some consulting has the balls to act like $great_prophet when proclaiming it. I mean, its not like Cablelabs hasn't been hard at work on the technologies to address the bandwidth issue. Both DOCSIS 3.0 (http://www.cablemodem.com/specifications/specific ations30.html) and Modular CMTS (http://www.cablemodem.com/specifications/m-cmts.h tml are designed to address this problem. M-CMTS basically works to divide cable plant into smaller sections by pushing the RF interfaces further out to the edge. This is done by placing fairly dumb/inexpensive edge QAM's out in the plant, these devices encapsulate DOCSIS frames into Gigabit Ethernet to carry them back to a packet processing engine. What this buys the operator is the ability to use fewer RF channels but gain more bandwidth at the cost of having some additional backhaul (to carry the GigE). Now some people might wonder if this consulting company is merely championing an idea that hasn't been developed, but sadly that isn't the case either. Many manufacturers are already producing EQAM's including big hitters like Cisco (http://www.cisco.com/en/US/products/hw/cable/ps22 09/products_implementation_design_guide_chapter091 86a00807c73c7.html/) These same EQAM's also handle switching of digital video so cable companies save on both switched video and normal IP traffic. DOCSIS 3.0 allows for bonding DOCSIS channels to create far more bandwidth, which is likely to be used for business services as well as more rich IP services. Comcast in my area already offers multiple HD on demand channels, for example HBO and Showtime. (http://www.comcast.com/HBOondemand/ and http://www.tvweek.com/news/2007/03/comcast_launche s_showtime_hdvo.php/)

Quite honestly it sounds like the "consultant" needs to do some research.

Amazingly there was only one intelligent thing said in the whole article. "Digital switching is key" is correct. Whats amazing is that some consulting has the balls to act like $great_prophet when proclaiming it. I mean, its not like Cablelabs hasn't been hard at work on the technologies to address the bandwidth issue. Both DOCSIS 3.0 (http://www.cablemodem.com/specifications/specific ations30.html) and Modular CMTS (http://www.cablemodem.com/specifications/m-cmts.h tml are designed to address this problem. M-CMTS basically works to divide cable plant into smaller sections by pushing the RF interfaces further out to the edge. This is done by placing fairly dumb/inexpensive edge QAM's out in the plant, these devices encapsulate DOCSIS frames into Gigabit Ethernet to carry them back to a packet processing engine. What this buys the operator is the ability to use fewer RF channels but gain more bandwidth at the cost of having some additional backhaul (to carry the GigE). Now some people might wonder if this consulting company is merely championing an idea that hasn't been developed, but sadly that isn't the case either. Many manufacturers are already producing EQAM's including big hitters like Cisco (http://www.cisco.com/en/US/products/hw/cable/ps22 09/products_implementation_design_guide_chapter091 86a00807c73c7.html/) These same EQAM's also handle switching of digital video so cable companies save on both switched video and normal IP traffic. DOCSIS 3.0 allows for bonding DOCSIS channels to create far more bandwidth, which is likely to be used for business services as well as more rich IP services. Comcast in my area already offers multiple HD on demand channels, for example HBO and Showtime. (http://www.comcast.com/HBOondemand/ and http://www.tvweek.com/news/2007/03/comcast_launche s_showtime_hdvo.php/)

Quite honestly it sounds like the "consultant" needs to do some research.

More information can be found at:
1. Specifications: http://www.cablemodem.com/specifications/specifica tions30.html
2. Press release: http://www.cablelabs.com/news/pr/2006/06_pr_docsis 30_080706.html
3. Ars Technica article: http://arstechnica.com/news.ars/post/20060808-7450 .html

Baseline Privacy Plus (BPI+) provides cable modem users with data privacy across the cable network. It does this by encrypting traffic flows between CM and CMTS.

Your "ahem features" are known as DOCSIS standards, which aren't unique to Adelphia.

Nor do they let us "snoop" any further than the modem itself, which is strictly diagnostic information.

so can most older proprietary modems. it has to do with encrypting traffic from the modem to the CMTS, which I suspect creates some overhead. perhaps they're just being cheap? I'm on Cox.net, and if I go to webmail.cox.net it's an http not an https on the page where you submit your username and password. On the public internet this would be an issue but inside the cox.net network, you can't sniff your neigbors traffic because of the way the modems are setup (no I haven't tried, but if I did, am fairly sure what I could see I couldn't read).

if you want all the dirt on how these modems work, go see the documentation at Cable Labs , they're the people who certify the equipment.

Thee's a reason I call myself broadbandbradley, I couldn't think of a good handle ;-)

Very true

All traffic coming in one side of a bridge will be sent out the other side.

This is only true for a simple bridge. I don't think any vendors currently provide simple bridges, since it is quite easy to incorporate the logic necessary to not forward every packet. From a bridge FAQ I found:

# Learning Bridges The simple bridges described above re-transmit every packet whether this is necessary or not. A learning bridge examines the source field of every packet it sees on each port and builds up a picture of which addresses are connected to which ports. This means that it will NOT re-transmit a packet if it knows that the destination address is connected to the same port as the bridge saw the packet on.

AFAIK, my cable modem (and all DOCSIS compliant cable modems) act in this manner.

But, all that applies to normal bridges. Your cable modem may actually be a router,

It is not a router, it does not use layer 3 (ip) addressing, it uses layer 2 (MAC) addressing. From the DOCSIS specification for external modems (section 2.1):
"The cable modem MUST be capable of filtering all broadcast traffic from the local LAN, with the exception of DHCP"
And from section 3.1.1.2.1:
"The cable modem MUST perform MAC bridging in accordance with ISO/IEC 10038 (ANSI/IEEE Std 802.1D)"

From the 802.1d specification section 7.1::
"The principal elements of bridge operation are:
a) Relay and filtering of frames (emphasis mine)

It would seem all DOCSIS 1.1 compliant cable modems are in fact learning bridges, and do not forward LAN traffic to the RF side of the modem. DSL "routers" often can act as a bridge or a router (or a combination 'brouter'), but I have never seen a cable modem that had those capabilites, all the equipment I encountered in @Home tech support was layer 2 bridging equipment. Of course, I certainly have not seen all possible equipment so YMMV.

And, as regards the bandwidth thing: your average DSL line gives you what? 1.5 Mbps? In one 6Mhz channel, cable can deliver 27Mpbs of bandwidth. There's a reason why most cable companies throttle bandwidth down to DSL speeds: it makes bandwidth much more consistent. Even without throttling, depending on your provider, it's highly unlikely that you'd be getting less than 512-1024 Kbps even during peak access hours. Compare that to DSL -- contrary to popular belief, everyone in your neighborhood is still sharing a single T3 trunk (sometimes less), so your neighbors can still impact your bandwidth during peak hours.

I mean, come on. On any network, when you have almost every host accessing data simultaneously, available bandwidth will drop. If the bottleneck isn't in the system itself, it's in the gateway out from there. Having used cable happily for several months now, I can say that I've never experienced these "slowdowns" that everyone talks about. And you can bet I'm on during "peak hours" -- isn't that about what time it is now (4:45 p.m. PDT)?

Ah, wait! I get it! You're nothing more than a cleverly disguised troll. My apologies, everyone. I won't feed him next time.

Not in this case. The 717MHz / 33MHz are the carrier frequencies used for the QPSK modulator /demodulator, but they do not represent the amount of bandwidth allocated to you. DOCSIS uses a combination of QPSK modulation and TDMA multiplexing because you've got several people transmitting / receiving on the same physical medium. So it's not the same thing as analog dial-up. You can read the DOCSIS 1.1 Radio Frequency Interface specification here

Unless you want to see how easy it is to produce convicing and very elaborate documentation of a fundamentally flawed exploit.

For those who won't bother reading the link (most of you), the exploit is this:

• DOCSIS Cable modems TFTP a file from the ISP to tell them what speed they are capped at (true)
• You can produce a docsis file (using the docsis project at sourceforge) that tells your cable modem to run at whatever speed you like (true).
• You can set the NIC IP on your PC to match the ISP's TFTP server, and set up your own TFTP server to serve your own docsis file (true).
• If you reset the cable modem, it will look on the PC side for the TFTP server, and user your docsis file (bzzzzt, false).

It looks really pretty until this last point, where it enters the realms of fantasy. The people who wrote the docsis spec aren't idiots. Cable modems will not look on the ethernet side for a TFTP server. TFTP'ing is done just after the cable side network discovery (so you have to have the cable side plugged in when you reset) and the modem knows which side is cable and which is ethernet. No, pinging the modem's ethernet IP from the PC doesn't help. It's just not that stupid; it knows that it has two interfaces, and it knows which one is which.

So go ahead and try this. You won't damage your modem, because it will simply ignore your TFTP server. What will happen is that you'll spend a couple of hours following the steps, getting all excited, then getting increasingly frustrated as you just can't get that last step to work. Rest assured, you're not doing anything wrong, other than following the instructions of a delusional wannabe hacker with a tiny amount of network knowledge and a real problem dealing with reality.

There are cable modem standards called DOCSIS. Not all modems follow these standards. For the modulation demodulation part have a look at the RF specification starting at section 4 on page 23. Upstream uses QPSK or 16QAM, downstream uses 64QAM or 256QAM. I believe that for a given number of homes there will be one downstream signal, but multiple upstream signals. (the number of upstreams would probably not equal the number of homes.)

For the hardcore RF geeks these specs are a great read. You can see how these specs were designed as opposed to "happened."

There are cable modem standards called DOCSIS. Not all modems follow these standards. For the modulation demodulation part have a look at the RF specification starting at section 4 on page 23. Upstream uses QPSK or 16QAM, downstream uses 64QAM or 256QAM. I believe that for a given number of homes there will be one downstream signal, but multiple upstream signals. (the number of upstreams would probably not equal the number of homes.)

For the hardcore RF geeks these specs are a great read. You can see how these specs were designed as opposed to "happened."

current cable modem technology based around the DOCSIS standard can give you up to 38Mbps downloads and 10 Mbps uploads. Cable companies currently cap that potential to about 1/10th of that. The Coax cable coming into your house has a 'bandwidth' of 750Mhz and each TV channel uses 6Mhz slots. Cable modems use 1 slot upstream and one slot downstream, meaning about 12Mhz out of the 750Mhz is used for your data connection. Digital TV boxes now can stuff about 10 TV channels into one 6Mhz slot. Obviously most of that 750Mhz is used for a broadcast medium.
Cable vendors can easily scale this approach to take away from 'broadcast' and move to 'download TV' where you could open up a guide and see what's on for the day and then pull a TIVO to download the show only if it's requested. in theory, this would allow them to dedicate more of that 750Mhz into Data-like connections and to provide that 38Mbps to anyone who wants it using the equipment currently installed in your home(if you already have a cable modem). think about 750Mhz divided into 6Mhz slots gives you 125 slots at a potential of 38Mbps per slot you come up with some 4,750Mbps downloading potential. Of course this approach would change the way people watch TV and fly in the face of traditional broadcast networks, but technology wise, the Cable providers are already there.
apllications? well I already stream my music in from the net at 128kbps, and downloading the latest Mozilla only takes a few minutes. Getting a copy of a new Linux distibution as ISO images (650Mb) still takes awhile.
still, I want More
currently I'd like more upstream to be able to do DV quality Video conferencing. I'd like more speed to be able to watch DVD quality video from the net like I stream my music today. DVD quality video can not yet be had with 100Mbps ethernet connection. I'd like to see them shoot for Gigabit ethernet to my house, I really need it.

PLEASE look up your facts before you post a message about a service you obviously don't know anything about!

The tech commissioned to install CableModems do not have to know anything other than 1) the RF level to the modem, 2) how to install a NIC (if the customer's not using USB or HomePNA), and 3) how to turn DHCP on in the computer.

DOCSIS CableModems go through a 'registration' process. I'll sum it up here:

The modem nabs and demodulates the downstream channel (the channel TO your cablemodem). In this, there will be information for its upstream, whereby it starts transmitting. Then it acquires a DHCP address on the 10 network. Next it finds a TFTP server and downloads its configuration and operational software (if there has been an update to this software compared to the version already in FLASH memory). It checks the TOD server and sets its time accordingly. Next up, it will bind with the MAC of your NIC and get an address for the CPE (your computer).

There are a lot of other things that happen, like ranging, SID assignment, and likewise.. You can find this information out just like I did by going to CableLabs and look through their DOCSIS documents.

Again, the TECHS do NOT have to configure the modems beyond the default configuration sent by the TFTP server. That is done by the engineers who set the system up, and set the system to default a certain configuration to the modem. They can change parameters specific to one modem remotely (the modem with this MAC gets this certain configuration the next time it registers) but you don't see technicians doing this work.

The cable companies are just starting to be service providers, and barely have the hang of it. Ask people who have had poor customer service from their broadband provider. You'll find them everywhere. That happens when they don't pay their employees! The 'good' people leave..

- Exigency

It depends on the modem. Lan City and Com21 proprietary modems are wide open. However the Docsis/Open Cable modems are (at least) bridge devices and can be configured to block ports. The really good docsis modems are routers (cisco) and do NAT. One of the cisco models will even support ppptp. for more info, check out the public access area of http://www.cablemodem.com, Baseline privacy document (.pdf file).