I am by no means a routing expert, but I believe there may be a fundamental flaw in your intentions.
I think you can provide redundent connectivity for outbound connections in the manner you describe, but for inbound, it is a different story. Basically, if you get a static IP from the Cable ISP and a different static IP from the DSL provider, then your inbound connections (for http or smtp or whatever) will be routed to the IP of either provider as configured by your DNS. If the cable company gives you x.x.x.x and the DSL is y.y.y.y and you configure your DNS as mail.yourdomain.com --> x.x.x.x When the circuit connecting x.x.x.x goes down, all of the servers trying to deliver mail will hold up until x.x.x.x comes back on-line. You will need to adjust DNS to get them to use y.y.y.y as your mail server. DNS changes propogate slowly. Too slowly. In the case of mail, you could setup y.y.y.y as a lower priority MX record, and that might work, but http and other protocols don't work that way. If your router is also serving as your mail server, it should properly handle which network card to send the reply packets out on (egress?). If your mail server is within your NAT domain, then you might consider setting up different NAT subnets for each of your ISP's and configure the mail server to use an IP alias on the same port for the two subnets. It could be seen by both connections then.... MX 10 --> x.x.x.x --> 192.168.10.111 --> your.mail.server and MX 20 --> y.y.y.y --> 192.168.20.111 --> your.mail.server (by aliases) I believe you are correct that BGP would solve your problem most properly, but is not an option. In that case, your routable addresses would change route when x.x.x.x went down. If email is critical, then you might consider using an email server that is external to your connection, like rent-a-redhat.com for $99 a month. Then your email connectivity becomes an outbound connection, which you can handle. The only trouble with this solution, is that a 2 MB attatchment going crom cubicle A to cubicle B must egress and ingress your DSL, which if it is ADSL, will be a bottleneck. I hope this helps.... On Tue, 4 Jun 2002, Rob Fegley wrote: > Hello! Please excuse if I've missed this topic elsewhere on this list, in the man >pages, or in a HOWTO somewhere. I'm about 2000 messages behind in my reading on this >list. I'll disclose right up front that most of my experience is in Cisco gear and >occasionally Alteon load-balancers, so excuse me if my questions seem a bit stupid or >if my expectations about how something should work in LEAF or Bering are contorted to >the Cisco world. Honestly, if I could run BGP with my Cable and DSL providers, I >wouldn't be posting any of the following questions. > > In any event, I have DSL already and will be accepting a cable modem "circuit" this >afternoon. I'm hoping to just toss another interface in my Bering box, and add >another default route out that interface. However, my questions are these: > > -With equal metrics assigned to two default routes, will traffic that ingresses on >one interface be routed back out of that same interface upon server reply, since I'm >port-forwarding inbound connections? This would imply that a port-forwarding >"session" table entry would take precedence over the routing table, right? This >would be my most preferred option, because it allows the greatest flexibility and >imparts the hardest work on Bering to figure out. > > -If not, then I need to apply a better cost to the interface that will do most of my >hosting, then apply some sort of periodic test that would flush my better cost >default route in the event that it's upstream path dies. The problem here is that >both interfaces will be plugging into a switch (on separate VLANs), but even if the >interfaces were crossover-cabled to my cable modem (bridge) and DSL bridge, the >Bering box should never see that interface link go down, so there is no route >flushing mechanism since a Layer 2 path always exists. Essentially, I am looking for >Bering to have some knowledge almost like a "hello timer" to some upstream device, >such that if visibility to that device (not necessarily another router, maybe my >ISP's DNS server) goes away, then a process kicks off to flush my current preferred >default route and uses the higher cost default. To read into this from a Cisco >perspective, I am looking for some method of simulating neighbor adjacency without p! eering with an upstream router, which is not an option. > > Both of the two previous questions are aimed at how the traffic flows back out to an >external client who made an initial inbound connection to something on my network. > > -Finally, in either an equal- or unequal-cost metric setup, does my outbound source >NAT (for my browsing) take place pre- or post-routing? In essence, by NATting my >internal subnet (or host) to an interface or an address within the address/netmask >applied to that interface, does that ensure that my traffic will egress on that same >interface, thus basically acting like policy routing? > > In closing, I appreciate *any* replies to these questions. Even if I get mostly >nasty replies would help as it would imply that I'm either asking too much out of one >box and may need to split this into pieces on several boxes, that I am totally in the >wrong place and need a different distribution, or need to pay for a piece of hardware >that is geared specifically to the tasks at hand. > > In any event, thank you for your time and consideration! > > Rob Fegley > TGI Micro > > _______________________________________________________________ > > Don't miss the 2002 Sprint PCS Application Developer's Conference > August 25-28 in Las Vegas -- http://devcon.sprintpcs.com/adp/index.cfm > > ------------------------------------------------------------------------ > leaf-user mailing list: [EMAIL PROTECTED] > https://lists.sourceforge.net/lists/listinfo/leaf-user > SR FAQ: http://leaf-project.org/pub/doc/docmanager/docid_1891.html > -- -------------------- Timothy Burt Internet Specialist _______________________________________________________________ Don't miss the 2002 Sprint PCS Application Developer's Conference August 25-28 in Las Vegas -- http://devcon.sprintpcs.com/adp/index.cfm ------------------------------------------------------------------------ leaf-user mailing list: [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/leaf-user SR FAQ: http://leaf-project.org/pub/doc/docmanager/docid_1891.html
