[HFC] pooling modems in layer2

Toney Mareo halflife4 at gmx.com
Thu Aug 14 12:48:58 UTC 2014


Thanks for the responses, I think it clarified a lot and I already started reading this CM-SP-L2VPN-I13-140403.pdf documentation.

What I need here is that existing clients are sent through ISP1 currently and I would like to add ISP2 for future clients without interfering anything with the current operations. Then later on move the old clients over to ISP2 as well.
As I see it, this can only be done on the CMTS device not after it unless it's possible to relay packets from the cable side with their original HFC macs through the CMTS.
Yes indeed I do not want to setup failover or balance DHCP servers, but I want to move every new subscriber to a different pool which gets directed to a different DHCP server which then finally able to provide the modems with ips and other settings to be able to go out on ISP2.

On Tue, Aug 12, 2014 at 10:23 AM, Toney Mareo <halflife4 at gmx.com> wrote:Hello

I think it's kind of an isp secret but I would be curious how do people distribute modems to pools before they would even reach the actual IP network so on layer2:

Certainly not secret, DOCSIS is a very well documented protocol with most of the information being publicly available.

For this I would like to get some clarification because I do not work in the telco industry. As I can figure out of the docsis, cablelabs documents. The CMTS device is connected to the coax segments through fiber. Therefore one could say that the "modem facing" side is a fiber optic interface but it's not 1000 Base-FX, not a regular Ethernet over fiber. It sends signals through a broad range of frequencies.
While fiber is commonly used in cable plants as part of a HFC network its completely transparent from a protocol standpoint the entire communication is over RF.  D3 and older uses QAM modulation and the downstream runs over "normal" 6 MHz channels which are the same as TV channels.
So what I would like to accomplish to provide a different pool of dhcp servers, which provides different config file, tod server, router, dns etc. infos to the modems but to do all this in Layer2.
Why?  The operator is the only one who can tell the CMTS which DHCP server(s) to send traffic to and modern CMTSs do that as an IP relay and passes its IP address as the GIADDR.
Because I advise the operator, you would think they are expert on the CMTS? Think again, I'm not an expert either but at least I learning.
I don't have hands on experience with CMTS-es but I would think that they are able to pool clients by MACs and able to send eg 500 clients to DHCP server1 and the other 1500 to DHCP server2 before they would even get an IP, so I talking of pure layer2 here!
Not exactly, first in nearly all cases the DHCP communication is an IP unicast rather than a layer 2 broadcast.  Second, the way that the DHCP server is selected is normally based on the type of device so that modems get a specific GIADDR, CPE (PCs, routers behind modems, etc) get another one, and often the EMTA gets a third.  It might be possible to do that off a count of devices, but if so it will be more of a load balancing scenario rather than these specific 500 CMs get this DHCP server.  It is possible to do open access in a DOCSIS system, but its very difficult and involves creating filters in both the CMTS and CM configurations.
Let's say if the CMTS device does not support this, what are the other options for routing layer2 traffic coming out of the CMTS? If I would know more about the device I would say that put a linuxbox after it (on the ISP facing nic) and mark the packets going out with arptables/ebtables then send them out of different nics to different dhcp servers.
It doesn't really work that way, but the closest thing is a "soft" tunnel that gets used for things like transparent LAN services, carrier WiFi, and a few other use cases. 
Any suggestions are welcome.

More information about the NANOG mailing list