Just wanted to reach out and get an idea how is people managing customers with static Ips, more specifically on Docsis networks where the customer could be moved between cmts's when a node is split
Thanks in advance for all responses,
Javier Gutierrez Guerra
Around here, Comcast seems to provision a GRE tunnel from the CPE back to some router within their network and run it over whatever IP address the CMTS hands out to the modem. Not very efficient, and it mandates that you use their CPE (they won't give you the necessary info to set it up yourself).
AFAIK, such service is only available on their "business class" DOCSIS product and is upcharged even then.
I do not believe it is a GRE tunnel.
I have seen (Charter) and heard quite a few run RIP or some other routing protocol on the CPE.
Though I have not seen anything specific about Comcast specifically.
Yep, it's RIP. They don't support IPv6 on this either. I've been asking for
IPv6 since 2006, it's always next year.
In my previous job, we managed static IPs by implementing L2TP tunnels between CPEs and central Juniper LNS. It wasn’t very elegant (at least for mtu, MSS clamping), but the design was done around 15 years ago
That's surprising to me, I have no intentions to do routing with our cable subscribers, that seems like a headache for both sides
Today we have specific ranges within subnets from where we assign IPs to customers, my main problem that I'm trying to get around is having to change a customer static IP if their node gets splitter and I have to mode them to a different CMTS
Javier Gutierrez Guerra
There’s really no good way to handle this without routing or tunneling that I’ve been able to find in a very long time. (SD-WAN can help, but it’s just a fancy way to tunnel in this regard.) It’s pretty amazing that this is such an issue, but it remains so. I have tried to work around this using BSoD (https://specification-search.cablelabs.com/business-services-over-docsis-layer-2-virtual-private-networks ) but we eventually abandoned the effort because it rapidly became to expensive to scale to solve a niche problem.
That's surprising to me, I have no intentions to do routing with our cable
subscribers, that seems like a headache for both sides
Meh, there are BNG solutions out there;
but RIP's not horrible _in_this_context_
Today we have specific ranges within subnets from where we assign IPs to
customers, my main problem that I'm trying to get around is having to
change a customer static IP if their node gets splitter and I have to mode
them to a different CMTS
I've seen some business services over DOCSIS where you're using VPLS to come
back to a central router of your choosing, but it's still Ethernet. There's
DHCP and mac addresses and a connection-less medium. PPPoE solves that, but
then you're doing PPP. Oh and you have the fun of doing ldp signaled VPLS
with a CMTS that was made by people who don't understand it.
There's several other BNG solutions out there, and I'm happy to go a bit
deeper if you're down the TR-69 path. For the business customer, most are
going to balk at anything other than a static IP Ethernet service delivered
over Ethernet. Depending on your skill-set and network size you might be able
to roll this yourself.
So in most cases I'm aware of, the cable provider did not use RIP directly
to a customer-managed device. The cable operator would deploy their own
managed device, implement RIP and the appropriate keychains between the
operator-managed premise device and the CMTS. As for the use cases, RIP was
implemented to address the specific 'recombine' use case where one day (or
evening) cable customer attachments could be moved from one CMTS to another.
Instrumenting that with DHCP or TR69 usually required other teams'
involvement and didn't allow portability.
With IPv6 you get PD which helps immensely.
On our network(which isn’t docsis, granted) we use PPPoE for all static IP addresses, because it allows /32 ip address allocations for all home CPE routers, upstream, the routers handle routing via ospf to change the path of where that /32 public IP goes. It allows “zero touch” moving of a customer from one PoP to another.
I am also interested in this. We are using DOCSIS and I am not sure what other providers with DOCSIS are using, any help on this will be appreciated.
Portable long prefix makes geoip more challenging (e.g. /29s in a
single /24 used in different markets)
I am curious for static IP services, how often does the address come
from a local (PoP/metro) pool, and what the policy is for keeping the
same IP when service address moves.
We have a provisioning system (promptlink) that we use to map cable modems
to their static ip addresses. The provisioning system has a gui front end
and it sits on linux and also acts as a dhcp server, etc. This is the same
ip address that we use for cable-helper (like ip-helper on a cmts bundle ip
interface) to forward dhcp requests from cable modem cpe, via the cmts, and
unicasted to promptlink and then the static ip address reservation within
the promptlink is sent back to the cpe
This all continues to work, even during node splits, as long as we don't
move that cm cpe to a different cmts... which would rarely happen since it's
across town to get to our other RF environment served be a different cmts
using a different static ip subnet... since we don't do L2 via cmts's in
order to stitch back that ip into a more globally located static subnet...
again, we don't do that. If the customers moves locations, into a different
cmts area, that would be required to give back the single static /32 ip and
get a different on. Unless they were a multi-static customer buying like a
/29... in which case we have no problem moving that /29 subnet off that cmts
and onto another one. That's easy.
We do however have more centrally located subnets for some of our single
static ip customers in FTTH... but not CMTS docsis.
Is it part of the DOCSIS spec that the CMTS terminates L3, or can they bridge to IEEE 802(.3) and delegate that to some other piece of gear? I'm unfortunately not familiar with the MSO world much at all aside from a little bit of L1.
The spec allows for bridging or layer 3 but none of the major or certified manufacturers support bridging on larger platforms. (>1000 modems)
I believe some of the bigger guys do BSoD (business services over DOCSI), L2 out the CMTS then take it to a central router for the CPEs gateway. Being that the GW lives on the CMTS in a normal setup, during node splits that would require customers moving to a new CMTS it might mean the customer doesn’t have access to that subnet anymore.
BSoD makes that a non-issue since their GW doesn't reside on the CMTS anymore, just move their vlan and they're good.
Aaron, I was thinking something similar. I've never once had a node split require moving a customer to a different CMTS. Even the very old and (relatively) low capacity 7200 VXR could serve several nodes per line card and supported several line cards per chassis. Newer cBR8, E6k, and the like can serve many many times more customers across dozens of nodes. Every L3 CMTS I've worked on uses something akin to ip unnumbered so as long as the customer stays on the same CMTS, their IP address will continue to work regardless of what interface or line card their connection terminates on.
E6K using gen 1 DCAMs can do about 32 service groups give or take, not that hard to get to a point with splits where you want to go past those numbers. Gen 2 DCAMs double that by going to 16 connectors compared to 8. cBR8 is less than the E6K.
The point of node splits is to lower customers per SG, you can’t just split and stay on the same chassis if you’re at capacity on slots.
If you take the Comcast approach and start pushing fiber deeper in order to remove actives your node counts sky rocket. All the whole they’re lowering counts on SGs as well.
Even us little guys or working on lowering customers per SG, they have to be moved somewhere which would be another chassis if you’re out of free connectors on like cards.
Sent from my iPhone
External Email: Use Caution
Aaron, I was thinking something similar. I’ve never once had a node
split require moving a customer to a different CMTS. Even the very old
and (relatively) low capacity 7200 VXR could serve several nodes per
line card and supported several line cards per chassis. Newer cBR8, E6k,
and the like can serve many many times more customers across dozens of
nodes. Every L3 CMTS I’ve worked on uses something akin to ip unnumbered
so as long as the customer stays on the same CMTS, their IP address will
continue to work regardless of what interface or line card their
connection terminates on.
16 connectors per DCAM2 times 6 cards is 96 DS service groups in a chassis. At ~1.2 Gbps per connector (using 32 SC-QAM DOCSIS 3.0 channels) that’s ~ 100gigabits per chassis. Quite a bit above my scale ;- )
The E6k can also do DOCSIS 3.1, which we use today, though I’m not sure what the capacity limit is per DCAM/SG/connector when both SQ-QAM and OFDM are used in combination.
You can do 32Q plus 2 OFDM blocks as well. But who has that kind of spectrum, we surely don’t. A 96Mhz block maybe.
But you can’t take the total at 100G and say that’s beyond your scale, you don’t run at full saturation do you?
And in order to run DCAM2s you’ll have to upgrade the RSMs to the gen2s as well.
Sent from my iPhone